xref: /openbmc/qemu/block.c (revision a89f364ae8740dfc31b321eed9ee454e996dc3c1)
1  /*
2   * QEMU System Emulator block driver
3   *
4   * Copyright (c) 2003 Fabrice Bellard
5   *
6   * Permission is hereby granted, free of charge, to any person obtaining a copy
7   * of this software and associated documentation files (the "Software"), to deal
8   * in the Software without restriction, including without limitation the rights
9   * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10   * copies of the Software, and to permit persons to whom the Software is
11   * furnished to do so, subject to the following conditions:
12   *
13   * The above copyright notice and this permission notice shall be included in
14   * all copies or substantial portions of the Software.
15   *
16   * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17   * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18   * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19   * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20   * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21   * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22   * THE SOFTWARE.
23   */
24  #include "qemu/osdep.h"
25  #include "block/trace.h"
26  #include "block/block_int.h"
27  #include "block/blockjob.h"
28  #include "block/nbd.h"
29  #include "qemu/error-report.h"
30  #include "module_block.h"
31  #include "qemu/module.h"
32  #include "qapi/qmp/qerror.h"
33  #include "qapi/qmp/qbool.h"
34  #include "qapi/qmp/qjson.h"
35  #include "sysemu/block-backend.h"
36  #include "sysemu/sysemu.h"
37  #include "qemu/notify.h"
38  #include "qemu/coroutine.h"
39  #include "block/qapi.h"
40  #include "qmp-commands.h"
41  #include "qemu/timer.h"
42  #include "qapi-event.h"
43  #include "qemu/cutils.h"
44  #include "qemu/id.h"
45  
46  #ifdef CONFIG_BSD
47  #include <sys/ioctl.h>
48  #include <sys/queue.h>
49  #ifndef __DragonFly__
50  #include <sys/disk.h>
51  #endif
52  #endif
53  
54  #ifdef _WIN32
55  #include <windows.h>
56  #endif
57  
58  #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
59  
60  static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
61      QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
62  
63  static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
64      QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
65  
66  static QLIST_HEAD(, BlockDriver) bdrv_drivers =
67      QLIST_HEAD_INITIALIZER(bdrv_drivers);
68  
69  static BlockDriverState *bdrv_open_inherit(const char *filename,
70                                             const char *reference,
71                                             QDict *options, int flags,
72                                             BlockDriverState *parent,
73                                             const BdrvChildRole *child_role,
74                                             Error **errp);
75  
76  /* If non-zero, use only whitelisted block drivers */
77  static int use_bdrv_whitelist;
78  
79  #ifdef _WIN32
80  static int is_windows_drive_prefix(const char *filename)
81  {
82      return (((filename[0] >= 'a' && filename[0] <= 'z') ||
83               (filename[0] >= 'A' && filename[0] <= 'Z')) &&
84              filename[1] == ':');
85  }
86  
87  int is_windows_drive(const char *filename)
88  {
89      if (is_windows_drive_prefix(filename) &&
90          filename[2] == '\0')
91          return 1;
92      if (strstart(filename, "\\\\.\\", NULL) ||
93          strstart(filename, "//./", NULL))
94          return 1;
95      return 0;
96  }
97  #endif
98  
99  size_t bdrv_opt_mem_align(BlockDriverState *bs)
100  {
101      if (!bs || !bs->drv) {
102          /* page size or 4k (hdd sector size) should be on the safe side */
103          return MAX(4096, getpagesize());
104      }
105  
106      return bs->bl.opt_mem_alignment;
107  }
108  
109  size_t bdrv_min_mem_align(BlockDriverState *bs)
110  {
111      if (!bs || !bs->drv) {
112          /* page size or 4k (hdd sector size) should be on the safe side */
113          return MAX(4096, getpagesize());
114      }
115  
116      return bs->bl.min_mem_alignment;
117  }
118  
119  /* check if the path starts with "<protocol>:" */
120  int path_has_protocol(const char *path)
121  {
122      const char *p;
123  
124  #ifdef _WIN32
125      if (is_windows_drive(path) ||
126          is_windows_drive_prefix(path)) {
127          return 0;
128      }
129      p = path + strcspn(path, ":/\\");
130  #else
131      p = path + strcspn(path, ":/");
132  #endif
133  
134      return *p == ':';
135  }
136  
137  int path_is_absolute(const char *path)
138  {
139  #ifdef _WIN32
140      /* specific case for names like: "\\.\d:" */
141      if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
142          return 1;
143      }
144      return (*path == '/' || *path == '\\');
145  #else
146      return (*path == '/');
147  #endif
148  }
149  
150  /* if filename is absolute, just copy it to dest. Otherwise, build a
151     path to it by considering it is relative to base_path. URL are
152     supported. */
153  void path_combine(char *dest, int dest_size,
154                    const char *base_path,
155                    const char *filename)
156  {
157      const char *p, *p1;
158      int len;
159  
160      if (dest_size <= 0)
161          return;
162      if (path_is_absolute(filename)) {
163          pstrcpy(dest, dest_size, filename);
164      } else {
165          const char *protocol_stripped = NULL;
166  
167          if (path_has_protocol(base_path)) {
168              protocol_stripped = strchr(base_path, ':');
169              if (protocol_stripped) {
170                  protocol_stripped++;
171              }
172          }
173          p = protocol_stripped ?: base_path;
174  
175          p1 = strrchr(base_path, '/');
176  #ifdef _WIN32
177          {
178              const char *p2;
179              p2 = strrchr(base_path, '\\');
180              if (!p1 || p2 > p1)
181                  p1 = p2;
182          }
183  #endif
184          if (p1)
185              p1++;
186          else
187              p1 = base_path;
188          if (p1 > p)
189              p = p1;
190          len = p - base_path;
191          if (len > dest_size - 1)
192              len = dest_size - 1;
193          memcpy(dest, base_path, len);
194          dest[len] = '\0';
195          pstrcat(dest, dest_size, filename);
196      }
197  }
198  
199  /*
200   * Helper function for bdrv_parse_filename() implementations to remove optional
201   * protocol prefixes (especially "file:") from a filename and for putting the
202   * stripped filename into the options QDict if there is such a prefix.
203   */
204  void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
205                                        QDict *options)
206  {
207      if (strstart(filename, prefix, &filename)) {
208          /* Stripping the explicit protocol prefix may result in a protocol
209           * prefix being (wrongly) detected (if the filename contains a colon) */
210          if (path_has_protocol(filename)) {
211              QString *fat_filename;
212  
213              /* This means there is some colon before the first slash; therefore,
214               * this cannot be an absolute path */
215              assert(!path_is_absolute(filename));
216  
217              /* And we can thus fix the protocol detection issue by prefixing it
218               * by "./" */
219              fat_filename = qstring_from_str("./");
220              qstring_append(fat_filename, filename);
221  
222              assert(!path_has_protocol(qstring_get_str(fat_filename)));
223  
224              qdict_put(options, "filename", fat_filename);
225          } else {
226              /* If no protocol prefix was detected, we can use the shortened
227               * filename as-is */
228              qdict_put_str(options, "filename", filename);
229          }
230      }
231  }
232  
233  
234  /* Returns whether the image file is opened as read-only. Note that this can
235   * return false and writing to the image file is still not possible because the
236   * image is inactivated. */
237  bool bdrv_is_read_only(BlockDriverState *bs)
238  {
239      return bs->read_only;
240  }
241  
242  int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
243                             bool ignore_allow_rdw, Error **errp)
244  {
245      /* Do not set read_only if copy_on_read is enabled */
246      if (bs->copy_on_read && read_only) {
247          error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
248                     bdrv_get_device_or_node_name(bs));
249          return -EINVAL;
250      }
251  
252      /* Do not clear read_only if it is prohibited */
253      if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
254          !ignore_allow_rdw)
255      {
256          error_setg(errp, "Node '%s' is read only",
257                     bdrv_get_device_or_node_name(bs));
258          return -EPERM;
259      }
260  
261      return 0;
262  }
263  
264  /* TODO Remove (deprecated since 2.11)
265   * Block drivers are not supposed to automatically change bs->read_only.
266   * Instead, they should just check whether they can provide what the user
267   * explicitly requested and error out if read-write is requested, but they can
268   * only provide read-only access. */
269  int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
270  {
271      int ret = 0;
272  
273      ret = bdrv_can_set_read_only(bs, read_only, false, errp);
274      if (ret < 0) {
275          return ret;
276      }
277  
278      bs->read_only = read_only;
279      return 0;
280  }
281  
282  void bdrv_get_full_backing_filename_from_filename(const char *backed,
283                                                    const char *backing,
284                                                    char *dest, size_t sz,
285                                                    Error **errp)
286  {
287      if (backing[0] == '\0' || path_has_protocol(backing) ||
288          path_is_absolute(backing))
289      {
290          pstrcpy(dest, sz, backing);
291      } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
292          error_setg(errp, "Cannot use relative backing file names for '%s'",
293                     backed);
294      } else {
295          path_combine(dest, sz, backed, backing);
296      }
297  }
298  
299  void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
300                                      Error **errp)
301  {
302      char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
303  
304      bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
305                                                   dest, sz, errp);
306  }
307  
308  void bdrv_register(BlockDriver *bdrv)
309  {
310      QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
311  }
312  
313  BlockDriverState *bdrv_new(void)
314  {
315      BlockDriverState *bs;
316      int i;
317  
318      bs = g_new0(BlockDriverState, 1);
319      QLIST_INIT(&bs->dirty_bitmaps);
320      for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
321          QLIST_INIT(&bs->op_blockers[i]);
322      }
323      notifier_with_return_list_init(&bs->before_write_notifiers);
324      qemu_co_mutex_init(&bs->reqs_lock);
325      qemu_mutex_init(&bs->dirty_bitmap_mutex);
326      bs->refcnt = 1;
327      bs->aio_context = qemu_get_aio_context();
328  
329      qemu_co_queue_init(&bs->flush_queue);
330  
331      QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
332  
333      return bs;
334  }
335  
336  static BlockDriver *bdrv_do_find_format(const char *format_name)
337  {
338      BlockDriver *drv1;
339  
340      QLIST_FOREACH(drv1, &bdrv_drivers, list) {
341          if (!strcmp(drv1->format_name, format_name)) {
342              return drv1;
343          }
344      }
345  
346      return NULL;
347  }
348  
349  BlockDriver *bdrv_find_format(const char *format_name)
350  {
351      BlockDriver *drv1;
352      int i;
353  
354      drv1 = bdrv_do_find_format(format_name);
355      if (drv1) {
356          return drv1;
357      }
358  
359      /* The driver isn't registered, maybe we need to load a module */
360      for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
361          if (!strcmp(block_driver_modules[i].format_name, format_name)) {
362              block_module_load_one(block_driver_modules[i].library_name);
363              break;
364          }
365      }
366  
367      return bdrv_do_find_format(format_name);
368  }
369  
370  static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
371  {
372      static const char *whitelist_rw[] = {
373          CONFIG_BDRV_RW_WHITELIST
374      };
375      static const char *whitelist_ro[] = {
376          CONFIG_BDRV_RO_WHITELIST
377      };
378      const char **p;
379  
380      if (!whitelist_rw[0] && !whitelist_ro[0]) {
381          return 1;               /* no whitelist, anything goes */
382      }
383  
384      for (p = whitelist_rw; *p; p++) {
385          if (!strcmp(drv->format_name, *p)) {
386              return 1;
387          }
388      }
389      if (read_only) {
390          for (p = whitelist_ro; *p; p++) {
391              if (!strcmp(drv->format_name, *p)) {
392                  return 1;
393              }
394          }
395      }
396      return 0;
397  }
398  
399  bool bdrv_uses_whitelist(void)
400  {
401      return use_bdrv_whitelist;
402  }
403  
404  typedef struct CreateCo {
405      BlockDriver *drv;
406      char *filename;
407      QemuOpts *opts;
408      int ret;
409      Error *err;
410  } CreateCo;
411  
412  static void coroutine_fn bdrv_create_co_entry(void *opaque)
413  {
414      Error *local_err = NULL;
415      int ret;
416  
417      CreateCo *cco = opaque;
418      assert(cco->drv);
419  
420      ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
421      error_propagate(&cco->err, local_err);
422      cco->ret = ret;
423  }
424  
425  int bdrv_create(BlockDriver *drv, const char* filename,
426                  QemuOpts *opts, Error **errp)
427  {
428      int ret;
429  
430      Coroutine *co;
431      CreateCo cco = {
432          .drv = drv,
433          .filename = g_strdup(filename),
434          .opts = opts,
435          .ret = NOT_DONE,
436          .err = NULL,
437      };
438  
439      if (!drv->bdrv_create) {
440          error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
441          ret = -ENOTSUP;
442          goto out;
443      }
444  
445      if (qemu_in_coroutine()) {
446          /* Fast-path if already in coroutine context */
447          bdrv_create_co_entry(&cco);
448      } else {
449          co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
450          qemu_coroutine_enter(co);
451          while (cco.ret == NOT_DONE) {
452              aio_poll(qemu_get_aio_context(), true);
453          }
454      }
455  
456      ret = cco.ret;
457      if (ret < 0) {
458          if (cco.err) {
459              error_propagate(errp, cco.err);
460          } else {
461              error_setg_errno(errp, -ret, "Could not create image");
462          }
463      }
464  
465  out:
466      g_free(cco.filename);
467      return ret;
468  }
469  
470  int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
471  {
472      BlockDriver *drv;
473      Error *local_err = NULL;
474      int ret;
475  
476      drv = bdrv_find_protocol(filename, true, errp);
477      if (drv == NULL) {
478          return -ENOENT;
479      }
480  
481      ret = bdrv_create(drv, filename, opts, &local_err);
482      error_propagate(errp, local_err);
483      return ret;
484  }
485  
486  /**
487   * Try to get @bs's logical and physical block size.
488   * On success, store them in @bsz struct and return 0.
489   * On failure return -errno.
490   * @bs must not be empty.
491   */
492  int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
493  {
494      BlockDriver *drv = bs->drv;
495  
496      if (drv && drv->bdrv_probe_blocksizes) {
497          return drv->bdrv_probe_blocksizes(bs, bsz);
498      } else if (drv && drv->is_filter && bs->file) {
499          return bdrv_probe_blocksizes(bs->file->bs, bsz);
500      }
501  
502      return -ENOTSUP;
503  }
504  
505  /**
506   * Try to get @bs's geometry (cyls, heads, sectors).
507   * On success, store them in @geo struct and return 0.
508   * On failure return -errno.
509   * @bs must not be empty.
510   */
511  int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
512  {
513      BlockDriver *drv = bs->drv;
514  
515      if (drv && drv->bdrv_probe_geometry) {
516          return drv->bdrv_probe_geometry(bs, geo);
517      } else if (drv && drv->is_filter && bs->file) {
518          return bdrv_probe_geometry(bs->file->bs, geo);
519      }
520  
521      return -ENOTSUP;
522  }
523  
524  /*
525   * Create a uniquely-named empty temporary file.
526   * Return 0 upon success, otherwise a negative errno value.
527   */
528  int get_tmp_filename(char *filename, int size)
529  {
530  #ifdef _WIN32
531      char temp_dir[MAX_PATH];
532      /* GetTempFileName requires that its output buffer (4th param)
533         have length MAX_PATH or greater.  */
534      assert(size >= MAX_PATH);
535      return (GetTempPath(MAX_PATH, temp_dir)
536              && GetTempFileName(temp_dir, "qem", 0, filename)
537              ? 0 : -GetLastError());
538  #else
539      int fd;
540      const char *tmpdir;
541      tmpdir = getenv("TMPDIR");
542      if (!tmpdir) {
543          tmpdir = "/var/tmp";
544      }
545      if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
546          return -EOVERFLOW;
547      }
548      fd = mkstemp(filename);
549      if (fd < 0) {
550          return -errno;
551      }
552      if (close(fd) != 0) {
553          unlink(filename);
554          return -errno;
555      }
556      return 0;
557  #endif
558  }
559  
560  /*
561   * Detect host devices. By convention, /dev/cdrom[N] is always
562   * recognized as a host CDROM.
563   */
564  static BlockDriver *find_hdev_driver(const char *filename)
565  {
566      int score_max = 0, score;
567      BlockDriver *drv = NULL, *d;
568  
569      QLIST_FOREACH(d, &bdrv_drivers, list) {
570          if (d->bdrv_probe_device) {
571              score = d->bdrv_probe_device(filename);
572              if (score > score_max) {
573                  score_max = score;
574                  drv = d;
575              }
576          }
577      }
578  
579      return drv;
580  }
581  
582  static BlockDriver *bdrv_do_find_protocol(const char *protocol)
583  {
584      BlockDriver *drv1;
585  
586      QLIST_FOREACH(drv1, &bdrv_drivers, list) {
587          if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
588              return drv1;
589          }
590      }
591  
592      return NULL;
593  }
594  
595  BlockDriver *bdrv_find_protocol(const char *filename,
596                                  bool allow_protocol_prefix,
597                                  Error **errp)
598  {
599      BlockDriver *drv1;
600      char protocol[128];
601      int len;
602      const char *p;
603      int i;
604  
605      /* TODO Drivers without bdrv_file_open must be specified explicitly */
606  
607      /*
608       * XXX(hch): we really should not let host device detection
609       * override an explicit protocol specification, but moving this
610       * later breaks access to device names with colons in them.
611       * Thanks to the brain-dead persistent naming schemes on udev-
612       * based Linux systems those actually are quite common.
613       */
614      drv1 = find_hdev_driver(filename);
615      if (drv1) {
616          return drv1;
617      }
618  
619      if (!path_has_protocol(filename) || !allow_protocol_prefix) {
620          return &bdrv_file;
621      }
622  
623      p = strchr(filename, ':');
624      assert(p != NULL);
625      len = p - filename;
626      if (len > sizeof(protocol) - 1)
627          len = sizeof(protocol) - 1;
628      memcpy(protocol, filename, len);
629      protocol[len] = '\0';
630  
631      drv1 = bdrv_do_find_protocol(protocol);
632      if (drv1) {
633          return drv1;
634      }
635  
636      for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
637          if (block_driver_modules[i].protocol_name &&
638              !strcmp(block_driver_modules[i].protocol_name, protocol)) {
639              block_module_load_one(block_driver_modules[i].library_name);
640              break;
641          }
642      }
643  
644      drv1 = bdrv_do_find_protocol(protocol);
645      if (!drv1) {
646          error_setg(errp, "Unknown protocol '%s'", protocol);
647      }
648      return drv1;
649  }
650  
651  /*
652   * Guess image format by probing its contents.
653   * This is not a good idea when your image is raw (CVE-2008-2004), but
654   * we do it anyway for backward compatibility.
655   *
656   * @buf         contains the image's first @buf_size bytes.
657   * @buf_size    is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
658   *              but can be smaller if the image file is smaller)
659   * @filename    is its filename.
660   *
661   * For all block drivers, call the bdrv_probe() method to get its
662   * probing score.
663   * Return the first block driver with the highest probing score.
664   */
665  BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
666                              const char *filename)
667  {
668      int score_max = 0, score;
669      BlockDriver *drv = NULL, *d;
670  
671      QLIST_FOREACH(d, &bdrv_drivers, list) {
672          if (d->bdrv_probe) {
673              score = d->bdrv_probe(buf, buf_size, filename);
674              if (score > score_max) {
675                  score_max = score;
676                  drv = d;
677              }
678          }
679      }
680  
681      return drv;
682  }
683  
684  static int find_image_format(BlockBackend *file, const char *filename,
685                               BlockDriver **pdrv, Error **errp)
686  {
687      BlockDriver *drv;
688      uint8_t buf[BLOCK_PROBE_BUF_SIZE];
689      int ret = 0;
690  
691      /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
692      if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
693          *pdrv = &bdrv_raw;
694          return ret;
695      }
696  
697      ret = blk_pread(file, 0, buf, sizeof(buf));
698      if (ret < 0) {
699          error_setg_errno(errp, -ret, "Could not read image for determining its "
700                           "format");
701          *pdrv = NULL;
702          return ret;
703      }
704  
705      drv = bdrv_probe_all(buf, ret, filename);
706      if (!drv) {
707          error_setg(errp, "Could not determine image format: No compatible "
708                     "driver found");
709          ret = -ENOENT;
710      }
711      *pdrv = drv;
712      return ret;
713  }
714  
715  /**
716   * Set the current 'total_sectors' value
717   * Return 0 on success, -errno on error.
718   */
719  static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
720  {
721      BlockDriver *drv = bs->drv;
722  
723      if (!drv) {
724          return -ENOMEDIUM;
725      }
726  
727      /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
728      if (bdrv_is_sg(bs))
729          return 0;
730  
731      /* query actual device if possible, otherwise just trust the hint */
732      if (drv->bdrv_getlength) {
733          int64_t length = drv->bdrv_getlength(bs);
734          if (length < 0) {
735              return length;
736          }
737          hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
738      }
739  
740      bs->total_sectors = hint;
741      return 0;
742  }
743  
744  /**
745   * Combines a QDict of new block driver @options with any missing options taken
746   * from @old_options, so that leaving out an option defaults to its old value.
747   */
748  static void bdrv_join_options(BlockDriverState *bs, QDict *options,
749                                QDict *old_options)
750  {
751      if (bs->drv && bs->drv->bdrv_join_options) {
752          bs->drv->bdrv_join_options(options, old_options);
753      } else {
754          qdict_join(options, old_options, false);
755      }
756  }
757  
758  /**
759   * Set open flags for a given discard mode
760   *
761   * Return 0 on success, -1 if the discard mode was invalid.
762   */
763  int bdrv_parse_discard_flags(const char *mode, int *flags)
764  {
765      *flags &= ~BDRV_O_UNMAP;
766  
767      if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
768          /* do nothing */
769      } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
770          *flags |= BDRV_O_UNMAP;
771      } else {
772          return -1;
773      }
774  
775      return 0;
776  }
777  
778  /**
779   * Set open flags for a given cache mode
780   *
781   * Return 0 on success, -1 if the cache mode was invalid.
782   */
783  int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
784  {
785      *flags &= ~BDRV_O_CACHE_MASK;
786  
787      if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
788          *writethrough = false;
789          *flags |= BDRV_O_NOCACHE;
790      } else if (!strcmp(mode, "directsync")) {
791          *writethrough = true;
792          *flags |= BDRV_O_NOCACHE;
793      } else if (!strcmp(mode, "writeback")) {
794          *writethrough = false;
795      } else if (!strcmp(mode, "unsafe")) {
796          *writethrough = false;
797          *flags |= BDRV_O_NO_FLUSH;
798      } else if (!strcmp(mode, "writethrough")) {
799          *writethrough = true;
800      } else {
801          return -1;
802      }
803  
804      return 0;
805  }
806  
807  static char *bdrv_child_get_parent_desc(BdrvChild *c)
808  {
809      BlockDriverState *parent = c->opaque;
810      return g_strdup(bdrv_get_device_or_node_name(parent));
811  }
812  
813  static void bdrv_child_cb_drained_begin(BdrvChild *child)
814  {
815      BlockDriverState *bs = child->opaque;
816      bdrv_drained_begin(bs);
817  }
818  
819  static void bdrv_child_cb_drained_end(BdrvChild *child)
820  {
821      BlockDriverState *bs = child->opaque;
822      bdrv_drained_end(bs);
823  }
824  
825  static void bdrv_child_cb_attach(BdrvChild *child)
826  {
827      BlockDriverState *bs = child->opaque;
828      bdrv_apply_subtree_drain(child, bs);
829  }
830  
831  static void bdrv_child_cb_detach(BdrvChild *child)
832  {
833      BlockDriverState *bs = child->opaque;
834      bdrv_unapply_subtree_drain(child, bs);
835  }
836  
837  static int bdrv_child_cb_inactivate(BdrvChild *child)
838  {
839      BlockDriverState *bs = child->opaque;
840      assert(bs->open_flags & BDRV_O_INACTIVE);
841      return 0;
842  }
843  
844  /*
845   * Returns the options and flags that a temporary snapshot should get, based on
846   * the originally requested flags (the originally requested image will have
847   * flags like a backing file)
848   */
849  static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
850                                         int parent_flags, QDict *parent_options)
851  {
852      *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
853  
854      /* For temporary files, unconditional cache=unsafe is fine */
855      qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
856      qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
857  
858      /* Copy the read-only option from the parent */
859      qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
860  
861      /* aio=native doesn't work for cache.direct=off, so disable it for the
862       * temporary snapshot */
863      *child_flags &= ~BDRV_O_NATIVE_AIO;
864  }
865  
866  /*
867   * Returns the options and flags that bs->file should get if a protocol driver
868   * is expected, based on the given options and flags for the parent BDS
869   */
870  static void bdrv_inherited_options(int *child_flags, QDict *child_options,
871                                     int parent_flags, QDict *parent_options)
872  {
873      int flags = parent_flags;
874  
875      /* Enable protocol handling, disable format probing for bs->file */
876      flags |= BDRV_O_PROTOCOL;
877  
878      /* If the cache mode isn't explicitly set, inherit direct and no-flush from
879       * the parent. */
880      qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
881      qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
882      qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
883  
884      /* Inherit the read-only option from the parent if it's not set */
885      qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
886  
887      /* Our block drivers take care to send flushes and respect unmap policy,
888       * so we can default to enable both on lower layers regardless of the
889       * corresponding parent options. */
890      qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
891  
892      /* Clear flags that only apply to the top layer */
893      flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
894                 BDRV_O_NO_IO);
895  
896      *child_flags = flags;
897  }
898  
899  const BdrvChildRole child_file = {
900      .get_parent_desc = bdrv_child_get_parent_desc,
901      .inherit_options = bdrv_inherited_options,
902      .drained_begin   = bdrv_child_cb_drained_begin,
903      .drained_end     = bdrv_child_cb_drained_end,
904      .attach          = bdrv_child_cb_attach,
905      .detach          = bdrv_child_cb_detach,
906      .inactivate      = bdrv_child_cb_inactivate,
907  };
908  
909  /*
910   * Returns the options and flags that bs->file should get if the use of formats
911   * (and not only protocols) is permitted for it, based on the given options and
912   * flags for the parent BDS
913   */
914  static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
915                                         int parent_flags, QDict *parent_options)
916  {
917      child_file.inherit_options(child_flags, child_options,
918                                 parent_flags, parent_options);
919  
920      *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
921  }
922  
923  const BdrvChildRole child_format = {
924      .get_parent_desc = bdrv_child_get_parent_desc,
925      .inherit_options = bdrv_inherited_fmt_options,
926      .drained_begin   = bdrv_child_cb_drained_begin,
927      .drained_end     = bdrv_child_cb_drained_end,
928      .attach          = bdrv_child_cb_attach,
929      .detach          = bdrv_child_cb_detach,
930      .inactivate      = bdrv_child_cb_inactivate,
931  };
932  
933  static void bdrv_backing_attach(BdrvChild *c)
934  {
935      BlockDriverState *parent = c->opaque;
936      BlockDriverState *backing_hd = c->bs;
937  
938      assert(!parent->backing_blocker);
939      error_setg(&parent->backing_blocker,
940                 "node is used as backing hd of '%s'",
941                 bdrv_get_device_or_node_name(parent));
942  
943      parent->open_flags &= ~BDRV_O_NO_BACKING;
944      pstrcpy(parent->backing_file, sizeof(parent->backing_file),
945              backing_hd->filename);
946      pstrcpy(parent->backing_format, sizeof(parent->backing_format),
947              backing_hd->drv ? backing_hd->drv->format_name : "");
948  
949      bdrv_op_block_all(backing_hd, parent->backing_blocker);
950      /* Otherwise we won't be able to commit or stream */
951      bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
952                      parent->backing_blocker);
953      bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
954                      parent->backing_blocker);
955      /*
956       * We do backup in 3 ways:
957       * 1. drive backup
958       *    The target bs is new opened, and the source is top BDS
959       * 2. blockdev backup
960       *    Both the source and the target are top BDSes.
961       * 3. internal backup(used for block replication)
962       *    Both the source and the target are backing file
963       *
964       * In case 1 and 2, neither the source nor the target is the backing file.
965       * In case 3, we will block the top BDS, so there is only one block job
966       * for the top BDS and its backing chain.
967       */
968      bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
969                      parent->backing_blocker);
970      bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
971                      parent->backing_blocker);
972  
973      bdrv_child_cb_attach(c);
974  }
975  
976  static void bdrv_backing_detach(BdrvChild *c)
977  {
978      BlockDriverState *parent = c->opaque;
979  
980      assert(parent->backing_blocker);
981      bdrv_op_unblock_all(c->bs, parent->backing_blocker);
982      error_free(parent->backing_blocker);
983      parent->backing_blocker = NULL;
984  
985      bdrv_child_cb_detach(c);
986  }
987  
988  /*
989   * Returns the options and flags that bs->backing should get, based on the
990   * given options and flags for the parent BDS
991   */
992  static void bdrv_backing_options(int *child_flags, QDict *child_options,
993                                   int parent_flags, QDict *parent_options)
994  {
995      int flags = parent_flags;
996  
997      /* The cache mode is inherited unmodified for backing files; except WCE,
998       * which is only applied on the top level (BlockBackend) */
999      qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1000      qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1001      qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1002  
1003      /* backing files always opened read-only */
1004      qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1005      flags &= ~BDRV_O_COPY_ON_READ;
1006  
1007      /* snapshot=on is handled on the top layer */
1008      flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1009  
1010      *child_flags = flags;
1011  }
1012  
1013  static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1014                                          const char *filename, Error **errp)
1015  {
1016      BlockDriverState *parent = c->opaque;
1017      int orig_flags = bdrv_get_flags(parent);
1018      int ret;
1019  
1020      if (!(orig_flags & BDRV_O_RDWR)) {
1021          ret = bdrv_reopen(parent, orig_flags | BDRV_O_RDWR, errp);
1022          if (ret < 0) {
1023              return ret;
1024          }
1025      }
1026  
1027      ret = bdrv_change_backing_file(parent, filename,
1028                                     base->drv ? base->drv->format_name : "");
1029      if (ret < 0) {
1030          error_setg_errno(errp, -ret, "Could not update backing file link");
1031      }
1032  
1033      if (!(orig_flags & BDRV_O_RDWR)) {
1034          bdrv_reopen(parent, orig_flags, NULL);
1035      }
1036  
1037      return ret;
1038  }
1039  
1040  const BdrvChildRole child_backing = {
1041      .get_parent_desc = bdrv_child_get_parent_desc,
1042      .attach          = bdrv_backing_attach,
1043      .detach          = bdrv_backing_detach,
1044      .inherit_options = bdrv_backing_options,
1045      .drained_begin   = bdrv_child_cb_drained_begin,
1046      .drained_end     = bdrv_child_cb_drained_end,
1047      .inactivate      = bdrv_child_cb_inactivate,
1048      .update_filename = bdrv_backing_update_filename,
1049  };
1050  
1051  static int bdrv_open_flags(BlockDriverState *bs, int flags)
1052  {
1053      int open_flags = flags;
1054  
1055      /*
1056       * Clear flags that are internal to the block layer before opening the
1057       * image.
1058       */
1059      open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1060  
1061      /*
1062       * Snapshots should be writable.
1063       */
1064      if (flags & BDRV_O_TEMPORARY) {
1065          open_flags |= BDRV_O_RDWR;
1066      }
1067  
1068      return open_flags;
1069  }
1070  
1071  static void update_flags_from_options(int *flags, QemuOpts *opts)
1072  {
1073      *flags &= ~BDRV_O_CACHE_MASK;
1074  
1075      assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
1076      if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1077          *flags |= BDRV_O_NO_FLUSH;
1078      }
1079  
1080      assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
1081      if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1082          *flags |= BDRV_O_NOCACHE;
1083      }
1084  
1085      *flags &= ~BDRV_O_RDWR;
1086  
1087      assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
1088      if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
1089          *flags |= BDRV_O_RDWR;
1090      }
1091  
1092  }
1093  
1094  static void update_options_from_flags(QDict *options, int flags)
1095  {
1096      if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1097          qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1098      }
1099      if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1100          qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1101                         flags & BDRV_O_NO_FLUSH);
1102      }
1103      if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1104          qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1105      }
1106  }
1107  
1108  static void bdrv_assign_node_name(BlockDriverState *bs,
1109                                    const char *node_name,
1110                                    Error **errp)
1111  {
1112      char *gen_node_name = NULL;
1113  
1114      if (!node_name) {
1115          node_name = gen_node_name = id_generate(ID_BLOCK);
1116      } else if (!id_wellformed(node_name)) {
1117          /*
1118           * Check for empty string or invalid characters, but not if it is
1119           * generated (generated names use characters not available to the user)
1120           */
1121          error_setg(errp, "Invalid node name");
1122          return;
1123      }
1124  
1125      /* takes care of avoiding namespaces collisions */
1126      if (blk_by_name(node_name)) {
1127          error_setg(errp, "node-name=%s is conflicting with a device id",
1128                     node_name);
1129          goto out;
1130      }
1131  
1132      /* takes care of avoiding duplicates node names */
1133      if (bdrv_find_node(node_name)) {
1134          error_setg(errp, "Duplicate node name");
1135          goto out;
1136      }
1137  
1138      /* copy node name into the bs and insert it into the graph list */
1139      pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1140      QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1141  out:
1142      g_free(gen_node_name);
1143  }
1144  
1145  static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1146                              const char *node_name, QDict *options,
1147                              int open_flags, Error **errp)
1148  {
1149      Error *local_err = NULL;
1150      int ret;
1151  
1152      bdrv_assign_node_name(bs, node_name, &local_err);
1153      if (local_err) {
1154          error_propagate(errp, local_err);
1155          return -EINVAL;
1156      }
1157  
1158      bs->drv = drv;
1159      bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1160      bs->opaque = g_malloc0(drv->instance_size);
1161  
1162      if (drv->bdrv_file_open) {
1163          assert(!drv->bdrv_needs_filename || bs->filename[0]);
1164          ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1165      } else if (drv->bdrv_open) {
1166          ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1167      } else {
1168          ret = 0;
1169      }
1170  
1171      if (ret < 0) {
1172          if (local_err) {
1173              error_propagate(errp, local_err);
1174          } else if (bs->filename[0]) {
1175              error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1176          } else {
1177              error_setg_errno(errp, -ret, "Could not open image");
1178          }
1179          goto open_failed;
1180      }
1181  
1182      ret = refresh_total_sectors(bs, bs->total_sectors);
1183      if (ret < 0) {
1184          error_setg_errno(errp, -ret, "Could not refresh total sector count");
1185          return ret;
1186      }
1187  
1188      bdrv_refresh_limits(bs, &local_err);
1189      if (local_err) {
1190          error_propagate(errp, local_err);
1191          return -EINVAL;
1192      }
1193  
1194      assert(bdrv_opt_mem_align(bs) != 0);
1195      assert(bdrv_min_mem_align(bs) != 0);
1196      assert(is_power_of_2(bs->bl.request_alignment));
1197  
1198      return 0;
1199  open_failed:
1200      bs->drv = NULL;
1201      if (bs->file != NULL) {
1202          bdrv_unref_child(bs, bs->file);
1203          bs->file = NULL;
1204      }
1205      g_free(bs->opaque);
1206      bs->opaque = NULL;
1207      return ret;
1208  }
1209  
1210  BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1211                                         int flags, Error **errp)
1212  {
1213      BlockDriverState *bs;
1214      int ret;
1215  
1216      bs = bdrv_new();
1217      bs->open_flags = flags;
1218      bs->explicit_options = qdict_new();
1219      bs->options = qdict_new();
1220      bs->opaque = NULL;
1221  
1222      update_options_from_flags(bs->options, flags);
1223  
1224      ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1225      if (ret < 0) {
1226          QDECREF(bs->explicit_options);
1227          bs->explicit_options = NULL;
1228          QDECREF(bs->options);
1229          bs->options = NULL;
1230          bdrv_unref(bs);
1231          return NULL;
1232      }
1233  
1234      return bs;
1235  }
1236  
1237  QemuOptsList bdrv_runtime_opts = {
1238      .name = "bdrv_common",
1239      .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1240      .desc = {
1241          {
1242              .name = "node-name",
1243              .type = QEMU_OPT_STRING,
1244              .help = "Node name of the block device node",
1245          },
1246          {
1247              .name = "driver",
1248              .type = QEMU_OPT_STRING,
1249              .help = "Block driver to use for the node",
1250          },
1251          {
1252              .name = BDRV_OPT_CACHE_DIRECT,
1253              .type = QEMU_OPT_BOOL,
1254              .help = "Bypass software writeback cache on the host",
1255          },
1256          {
1257              .name = BDRV_OPT_CACHE_NO_FLUSH,
1258              .type = QEMU_OPT_BOOL,
1259              .help = "Ignore flush requests",
1260          },
1261          {
1262              .name = BDRV_OPT_READ_ONLY,
1263              .type = QEMU_OPT_BOOL,
1264              .help = "Node is opened in read-only mode",
1265          },
1266          {
1267              .name = "detect-zeroes",
1268              .type = QEMU_OPT_STRING,
1269              .help = "try to optimize zero writes (off, on, unmap)",
1270          },
1271          {
1272              .name = "discard",
1273              .type = QEMU_OPT_STRING,
1274              .help = "discard operation (ignore/off, unmap/on)",
1275          },
1276          {
1277              .name = BDRV_OPT_FORCE_SHARE,
1278              .type = QEMU_OPT_BOOL,
1279              .help = "always accept other writers (default: off)",
1280          },
1281          { /* end of list */ }
1282      },
1283  };
1284  
1285  /*
1286   * Common part for opening disk images and files
1287   *
1288   * Removes all processed options from *options.
1289   */
1290  static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1291                              QDict *options, Error **errp)
1292  {
1293      int ret, open_flags;
1294      const char *filename;
1295      const char *driver_name = NULL;
1296      const char *node_name = NULL;
1297      const char *discard;
1298      const char *detect_zeroes;
1299      QemuOpts *opts;
1300      BlockDriver *drv;
1301      Error *local_err = NULL;
1302  
1303      assert(bs->file == NULL);
1304      assert(options != NULL && bs->options != options);
1305  
1306      opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1307      qemu_opts_absorb_qdict(opts, options, &local_err);
1308      if (local_err) {
1309          error_propagate(errp, local_err);
1310          ret = -EINVAL;
1311          goto fail_opts;
1312      }
1313  
1314      update_flags_from_options(&bs->open_flags, opts);
1315  
1316      driver_name = qemu_opt_get(opts, "driver");
1317      drv = bdrv_find_format(driver_name);
1318      assert(drv != NULL);
1319  
1320      bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1321  
1322      if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1323          error_setg(errp,
1324                     BDRV_OPT_FORCE_SHARE
1325                     "=on can only be used with read-only images");
1326          ret = -EINVAL;
1327          goto fail_opts;
1328      }
1329  
1330      if (file != NULL) {
1331          filename = blk_bs(file)->filename;
1332      } else {
1333          /*
1334           * Caution: while qdict_get_try_str() is fine, getting
1335           * non-string types would require more care.  When @options
1336           * come from -blockdev or blockdev_add, its members are typed
1337           * according to the QAPI schema, but when they come from
1338           * -drive, they're all QString.
1339           */
1340          filename = qdict_get_try_str(options, "filename");
1341      }
1342  
1343      if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1344          error_setg(errp, "The '%s' block driver requires a file name",
1345                     drv->format_name);
1346          ret = -EINVAL;
1347          goto fail_opts;
1348      }
1349  
1350      trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1351                             drv->format_name);
1352  
1353      bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1354  
1355      if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1356          error_setg(errp,
1357                     !bs->read_only && bdrv_is_whitelisted(drv, true)
1358                          ? "Driver '%s' can only be used for read-only devices"
1359                          : "Driver '%s' is not whitelisted",
1360                     drv->format_name);
1361          ret = -ENOTSUP;
1362          goto fail_opts;
1363      }
1364  
1365      /* bdrv_new() and bdrv_close() make it so */
1366      assert(atomic_read(&bs->copy_on_read) == 0);
1367  
1368      if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1369          if (!bs->read_only) {
1370              bdrv_enable_copy_on_read(bs);
1371          } else {
1372              error_setg(errp, "Can't use copy-on-read on read-only device");
1373              ret = -EINVAL;
1374              goto fail_opts;
1375          }
1376      }
1377  
1378      discard = qemu_opt_get(opts, "discard");
1379      if (discard != NULL) {
1380          if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1381              error_setg(errp, "Invalid discard option");
1382              ret = -EINVAL;
1383              goto fail_opts;
1384          }
1385      }
1386  
1387      detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1388      if (detect_zeroes) {
1389          BlockdevDetectZeroesOptions value =
1390              qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
1391                              detect_zeroes,
1392                              BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1393                              &local_err);
1394          if (local_err) {
1395              error_propagate(errp, local_err);
1396              ret = -EINVAL;
1397              goto fail_opts;
1398          }
1399  
1400          if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1401              !(bs->open_flags & BDRV_O_UNMAP))
1402          {
1403              error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1404                               "without setting discard operation to unmap");
1405              ret = -EINVAL;
1406              goto fail_opts;
1407          }
1408  
1409          bs->detect_zeroes = value;
1410      }
1411  
1412      if (filename != NULL) {
1413          pstrcpy(bs->filename, sizeof(bs->filename), filename);
1414      } else {
1415          bs->filename[0] = '\0';
1416      }
1417      pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1418  
1419      /* Open the image, either directly or using a protocol */
1420      open_flags = bdrv_open_flags(bs, bs->open_flags);
1421      node_name = qemu_opt_get(opts, "node-name");
1422  
1423      assert(!drv->bdrv_file_open || file == NULL);
1424      ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1425      if (ret < 0) {
1426          goto fail_opts;
1427      }
1428  
1429      qemu_opts_del(opts);
1430      return 0;
1431  
1432  fail_opts:
1433      qemu_opts_del(opts);
1434      return ret;
1435  }
1436  
1437  static QDict *parse_json_filename(const char *filename, Error **errp)
1438  {
1439      QObject *options_obj;
1440      QDict *options;
1441      int ret;
1442  
1443      ret = strstart(filename, "json:", &filename);
1444      assert(ret);
1445  
1446      options_obj = qobject_from_json(filename, errp);
1447      if (!options_obj) {
1448          /* Work around qobject_from_json() lossage TODO fix that */
1449          if (errp && !*errp) {
1450              error_setg(errp, "Could not parse the JSON options");
1451              return NULL;
1452          }
1453          error_prepend(errp, "Could not parse the JSON options: ");
1454          return NULL;
1455      }
1456  
1457      options = qobject_to_qdict(options_obj);
1458      if (!options) {
1459          qobject_decref(options_obj);
1460          error_setg(errp, "Invalid JSON object given");
1461          return NULL;
1462      }
1463  
1464      qdict_flatten(options);
1465  
1466      return options;
1467  }
1468  
1469  static void parse_json_protocol(QDict *options, const char **pfilename,
1470                                  Error **errp)
1471  {
1472      QDict *json_options;
1473      Error *local_err = NULL;
1474  
1475      /* Parse json: pseudo-protocol */
1476      if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1477          return;
1478      }
1479  
1480      json_options = parse_json_filename(*pfilename, &local_err);
1481      if (local_err) {
1482          error_propagate(errp, local_err);
1483          return;
1484      }
1485  
1486      /* Options given in the filename have lower priority than options
1487       * specified directly */
1488      qdict_join(options, json_options, false);
1489      QDECREF(json_options);
1490      *pfilename = NULL;
1491  }
1492  
1493  /*
1494   * Fills in default options for opening images and converts the legacy
1495   * filename/flags pair to option QDict entries.
1496   * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1497   * block driver has been specified explicitly.
1498   */
1499  static int bdrv_fill_options(QDict **options, const char *filename,
1500                               int *flags, Error **errp)
1501  {
1502      const char *drvname;
1503      bool protocol = *flags & BDRV_O_PROTOCOL;
1504      bool parse_filename = false;
1505      BlockDriver *drv = NULL;
1506      Error *local_err = NULL;
1507  
1508      /*
1509       * Caution: while qdict_get_try_str() is fine, getting non-string
1510       * types would require more care.  When @options come from
1511       * -blockdev or blockdev_add, its members are typed according to
1512       * the QAPI schema, but when they come from -drive, they're all
1513       * QString.
1514       */
1515      drvname = qdict_get_try_str(*options, "driver");
1516      if (drvname) {
1517          drv = bdrv_find_format(drvname);
1518          if (!drv) {
1519              error_setg(errp, "Unknown driver '%s'", drvname);
1520              return -ENOENT;
1521          }
1522          /* If the user has explicitly specified the driver, this choice should
1523           * override the BDRV_O_PROTOCOL flag */
1524          protocol = drv->bdrv_file_open;
1525      }
1526  
1527      if (protocol) {
1528          *flags |= BDRV_O_PROTOCOL;
1529      } else {
1530          *flags &= ~BDRV_O_PROTOCOL;
1531      }
1532  
1533      /* Translate cache options from flags into options */
1534      update_options_from_flags(*options, *flags);
1535  
1536      /* Fetch the file name from the options QDict if necessary */
1537      if (protocol && filename) {
1538          if (!qdict_haskey(*options, "filename")) {
1539              qdict_put_str(*options, "filename", filename);
1540              parse_filename = true;
1541          } else {
1542              error_setg(errp, "Can't specify 'file' and 'filename' options at "
1543                               "the same time");
1544              return -EINVAL;
1545          }
1546      }
1547  
1548      /* Find the right block driver */
1549      /* See cautionary note on accessing @options above */
1550      filename = qdict_get_try_str(*options, "filename");
1551  
1552      if (!drvname && protocol) {
1553          if (filename) {
1554              drv = bdrv_find_protocol(filename, parse_filename, errp);
1555              if (!drv) {
1556                  return -EINVAL;
1557              }
1558  
1559              drvname = drv->format_name;
1560              qdict_put_str(*options, "driver", drvname);
1561          } else {
1562              error_setg(errp, "Must specify either driver or file");
1563              return -EINVAL;
1564          }
1565      }
1566  
1567      assert(drv || !protocol);
1568  
1569      /* Driver-specific filename parsing */
1570      if (drv && drv->bdrv_parse_filename && parse_filename) {
1571          drv->bdrv_parse_filename(filename, *options, &local_err);
1572          if (local_err) {
1573              error_propagate(errp, local_err);
1574              return -EINVAL;
1575          }
1576  
1577          if (!drv->bdrv_needs_filename) {
1578              qdict_del(*options, "filename");
1579          }
1580      }
1581  
1582      return 0;
1583  }
1584  
1585  static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1586                                   uint64_t perm, uint64_t shared,
1587                                   GSList *ignore_children, Error **errp);
1588  static void bdrv_child_abort_perm_update(BdrvChild *c);
1589  static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1590  
1591  typedef struct BlockReopenQueueEntry {
1592       bool prepared;
1593       BDRVReopenState state;
1594       QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1595  } BlockReopenQueueEntry;
1596  
1597  /*
1598   * Return the flags that @bs will have after the reopens in @q have
1599   * successfully completed. If @q is NULL (or @bs is not contained in @q),
1600   * return the current flags.
1601   */
1602  static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1603  {
1604      BlockReopenQueueEntry *entry;
1605  
1606      if (q != NULL) {
1607          QSIMPLEQ_FOREACH(entry, q, entry) {
1608              if (entry->state.bs == bs) {
1609                  return entry->state.flags;
1610              }
1611          }
1612      }
1613  
1614      return bs->open_flags;
1615  }
1616  
1617  /* Returns whether the image file can be written to after the reopen queue @q
1618   * has been successfully applied, or right now if @q is NULL. */
1619  static bool bdrv_is_writable(BlockDriverState *bs, BlockReopenQueue *q)
1620  {
1621      int flags = bdrv_reopen_get_flags(q, bs);
1622  
1623      return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1624  }
1625  
1626  static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1627                              BdrvChild *c, const BdrvChildRole *role,
1628                              BlockReopenQueue *reopen_queue,
1629                              uint64_t parent_perm, uint64_t parent_shared,
1630                              uint64_t *nperm, uint64_t *nshared)
1631  {
1632      if (bs->drv && bs->drv->bdrv_child_perm) {
1633          bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1634                                   parent_perm, parent_shared,
1635                                   nperm, nshared);
1636      }
1637      /* TODO Take force_share from reopen_queue */
1638      if (child_bs && child_bs->force_share) {
1639          *nshared = BLK_PERM_ALL;
1640      }
1641  }
1642  
1643  /*
1644   * Check whether permissions on this node can be changed in a way that
1645   * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1646   * permissions of all its parents. This involves checking whether all necessary
1647   * permission changes to child nodes can be performed.
1648   *
1649   * A call to this function must always be followed by a call to bdrv_set_perm()
1650   * or bdrv_abort_perm_update().
1651   */
1652  static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1653                             uint64_t cumulative_perms,
1654                             uint64_t cumulative_shared_perms,
1655                             GSList *ignore_children, Error **errp)
1656  {
1657      BlockDriver *drv = bs->drv;
1658      BdrvChild *c;
1659      int ret;
1660  
1661      /* Write permissions never work with read-only images */
1662      if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1663          !bdrv_is_writable(bs, q))
1664      {
1665          error_setg(errp, "Block node is read-only");
1666          return -EPERM;
1667      }
1668  
1669      /* Check this node */
1670      if (!drv) {
1671          return 0;
1672      }
1673  
1674      if (drv->bdrv_check_perm) {
1675          return drv->bdrv_check_perm(bs, cumulative_perms,
1676                                      cumulative_shared_perms, errp);
1677      }
1678  
1679      /* Drivers that never have children can omit .bdrv_child_perm() */
1680      if (!drv->bdrv_child_perm) {
1681          assert(QLIST_EMPTY(&bs->children));
1682          return 0;
1683      }
1684  
1685      /* Check all children */
1686      QLIST_FOREACH(c, &bs->children, next) {
1687          uint64_t cur_perm, cur_shared;
1688          bdrv_child_perm(bs, c->bs, c, c->role, q,
1689                          cumulative_perms, cumulative_shared_perms,
1690                          &cur_perm, &cur_shared);
1691          ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1692                                      ignore_children, errp);
1693          if (ret < 0) {
1694              return ret;
1695          }
1696      }
1697  
1698      return 0;
1699  }
1700  
1701  /*
1702   * Notifies drivers that after a previous bdrv_check_perm() call, the
1703   * permission update is not performed and any preparations made for it (e.g.
1704   * taken file locks) need to be undone.
1705   *
1706   * This function recursively notifies all child nodes.
1707   */
1708  static void bdrv_abort_perm_update(BlockDriverState *bs)
1709  {
1710      BlockDriver *drv = bs->drv;
1711      BdrvChild *c;
1712  
1713      if (!drv) {
1714          return;
1715      }
1716  
1717      if (drv->bdrv_abort_perm_update) {
1718          drv->bdrv_abort_perm_update(bs);
1719      }
1720  
1721      QLIST_FOREACH(c, &bs->children, next) {
1722          bdrv_child_abort_perm_update(c);
1723      }
1724  }
1725  
1726  static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1727                            uint64_t cumulative_shared_perms)
1728  {
1729      BlockDriver *drv = bs->drv;
1730      BdrvChild *c;
1731  
1732      if (!drv) {
1733          return;
1734      }
1735  
1736      /* Update this node */
1737      if (drv->bdrv_set_perm) {
1738          drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1739      }
1740  
1741      /* Drivers that never have children can omit .bdrv_child_perm() */
1742      if (!drv->bdrv_child_perm) {
1743          assert(QLIST_EMPTY(&bs->children));
1744          return;
1745      }
1746  
1747      /* Update all children */
1748      QLIST_FOREACH(c, &bs->children, next) {
1749          uint64_t cur_perm, cur_shared;
1750          bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1751                          cumulative_perms, cumulative_shared_perms,
1752                          &cur_perm, &cur_shared);
1753          bdrv_child_set_perm(c, cur_perm, cur_shared);
1754      }
1755  }
1756  
1757  static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1758                                       uint64_t *shared_perm)
1759  {
1760      BdrvChild *c;
1761      uint64_t cumulative_perms = 0;
1762      uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1763  
1764      QLIST_FOREACH(c, &bs->parents, next_parent) {
1765          cumulative_perms |= c->perm;
1766          cumulative_shared_perms &= c->shared_perm;
1767      }
1768  
1769      *perm = cumulative_perms;
1770      *shared_perm = cumulative_shared_perms;
1771  }
1772  
1773  static char *bdrv_child_user_desc(BdrvChild *c)
1774  {
1775      if (c->role->get_parent_desc) {
1776          return c->role->get_parent_desc(c);
1777      }
1778  
1779      return g_strdup("another user");
1780  }
1781  
1782  char *bdrv_perm_names(uint64_t perm)
1783  {
1784      struct perm_name {
1785          uint64_t perm;
1786          const char *name;
1787      } permissions[] = {
1788          { BLK_PERM_CONSISTENT_READ, "consistent read" },
1789          { BLK_PERM_WRITE,           "write" },
1790          { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1791          { BLK_PERM_RESIZE,          "resize" },
1792          { BLK_PERM_GRAPH_MOD,       "change children" },
1793          { 0, NULL }
1794      };
1795  
1796      char *result = g_strdup("");
1797      struct perm_name *p;
1798  
1799      for (p = permissions; p->name; p++) {
1800          if (perm & p->perm) {
1801              char *old = result;
1802              result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1803              g_free(old);
1804          }
1805      }
1806  
1807      return result;
1808  }
1809  
1810  /*
1811   * Checks whether a new reference to @bs can be added if the new user requires
1812   * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1813   * set, the BdrvChild objects in this list are ignored in the calculations;
1814   * this allows checking permission updates for an existing reference.
1815   *
1816   * Needs to be followed by a call to either bdrv_set_perm() or
1817   * bdrv_abort_perm_update(). */
1818  static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1819                                    uint64_t new_used_perm,
1820                                    uint64_t new_shared_perm,
1821                                    GSList *ignore_children, Error **errp)
1822  {
1823      BdrvChild *c;
1824      uint64_t cumulative_perms = new_used_perm;
1825      uint64_t cumulative_shared_perms = new_shared_perm;
1826  
1827      /* There is no reason why anyone couldn't tolerate write_unchanged */
1828      assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1829  
1830      QLIST_FOREACH(c, &bs->parents, next_parent) {
1831          if (g_slist_find(ignore_children, c)) {
1832              continue;
1833          }
1834  
1835          if ((new_used_perm & c->shared_perm) != new_used_perm) {
1836              char *user = bdrv_child_user_desc(c);
1837              char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1838              error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1839                               "allow '%s' on %s",
1840                         user, c->name, perm_names, bdrv_get_node_name(c->bs));
1841              g_free(user);
1842              g_free(perm_names);
1843              return -EPERM;
1844          }
1845  
1846          if ((c->perm & new_shared_perm) != c->perm) {
1847              char *user = bdrv_child_user_desc(c);
1848              char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1849              error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1850                               "'%s' on %s",
1851                         user, c->name, perm_names, bdrv_get_node_name(c->bs));
1852              g_free(user);
1853              g_free(perm_names);
1854              return -EPERM;
1855          }
1856  
1857          cumulative_perms |= c->perm;
1858          cumulative_shared_perms &= c->shared_perm;
1859      }
1860  
1861      return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
1862                             ignore_children, errp);
1863  }
1864  
1865  /* Needs to be followed by a call to either bdrv_child_set_perm() or
1866   * bdrv_child_abort_perm_update(). */
1867  static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1868                                   uint64_t perm, uint64_t shared,
1869                                   GSList *ignore_children, Error **errp)
1870  {
1871      int ret;
1872  
1873      ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1874      ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
1875      g_slist_free(ignore_children);
1876  
1877      return ret;
1878  }
1879  
1880  static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1881  {
1882      uint64_t cumulative_perms, cumulative_shared_perms;
1883  
1884      c->perm = perm;
1885      c->shared_perm = shared;
1886  
1887      bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1888                               &cumulative_shared_perms);
1889      bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1890  }
1891  
1892  static void bdrv_child_abort_perm_update(BdrvChild *c)
1893  {
1894      bdrv_abort_perm_update(c->bs);
1895  }
1896  
1897  int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1898                              Error **errp)
1899  {
1900      int ret;
1901  
1902      ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
1903      if (ret < 0) {
1904          bdrv_child_abort_perm_update(c);
1905          return ret;
1906      }
1907  
1908      bdrv_child_set_perm(c, perm, shared);
1909  
1910      return 0;
1911  }
1912  
1913  #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1914                                   | BLK_PERM_WRITE \
1915                                   | BLK_PERM_WRITE_UNCHANGED \
1916                                   | BLK_PERM_RESIZE)
1917  #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1918  
1919  void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1920                                 const BdrvChildRole *role,
1921                                 BlockReopenQueue *reopen_queue,
1922                                 uint64_t perm, uint64_t shared,
1923                                 uint64_t *nperm, uint64_t *nshared)
1924  {
1925      if (c == NULL) {
1926          *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1927          *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1928          return;
1929      }
1930  
1931      *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1932               (c->perm & DEFAULT_PERM_UNCHANGED);
1933      *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1934                 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1935  }
1936  
1937  void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1938                                 const BdrvChildRole *role,
1939                                 BlockReopenQueue *reopen_queue,
1940                                 uint64_t perm, uint64_t shared,
1941                                 uint64_t *nperm, uint64_t *nshared)
1942  {
1943      bool backing = (role == &child_backing);
1944      assert(role == &child_backing || role == &child_file);
1945  
1946      if (!backing) {
1947          int flags = bdrv_reopen_get_flags(reopen_queue, bs);
1948  
1949          /* Apart from the modifications below, the same permissions are
1950           * forwarded and left alone as for filters */
1951          bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
1952                                    &perm, &shared);
1953  
1954          /* Format drivers may touch metadata even if the guest doesn't write */
1955          if (bdrv_is_writable(bs, reopen_queue)) {
1956              perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1957          }
1958  
1959          /* bs->file always needs to be consistent because of the metadata. We
1960           * can never allow other users to resize or write to it. */
1961          if (!(flags & BDRV_O_NO_IO)) {
1962              perm |= BLK_PERM_CONSISTENT_READ;
1963          }
1964          shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1965      } else {
1966          /* We want consistent read from backing files if the parent needs it.
1967           * No other operations are performed on backing files. */
1968          perm &= BLK_PERM_CONSISTENT_READ;
1969  
1970          /* If the parent can deal with changing data, we're okay with a
1971           * writable and resizable backing file. */
1972          /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1973          if (shared & BLK_PERM_WRITE) {
1974              shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1975          } else {
1976              shared = 0;
1977          }
1978  
1979          shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1980                    BLK_PERM_WRITE_UNCHANGED;
1981      }
1982  
1983      if (bs->open_flags & BDRV_O_INACTIVE) {
1984          shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1985      }
1986  
1987      *nperm = perm;
1988      *nshared = shared;
1989  }
1990  
1991  static void bdrv_replace_child_noperm(BdrvChild *child,
1992                                        BlockDriverState *new_bs)
1993  {
1994      BlockDriverState *old_bs = child->bs;
1995      int i;
1996  
1997      if (old_bs && new_bs) {
1998          assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
1999      }
2000      if (old_bs) {
2001          /* Detach first so that the recursive drain sections coming from @child
2002           * are already gone and we only end the drain sections that came from
2003           * elsewhere. */
2004          if (child->role->detach) {
2005              child->role->detach(child);
2006          }
2007          if (old_bs->quiesce_counter && child->role->drained_end) {
2008              for (i = 0; i < old_bs->quiesce_counter; i++) {
2009                  child->role->drained_end(child);
2010              }
2011          }
2012          QLIST_REMOVE(child, next_parent);
2013      }
2014  
2015      child->bs = new_bs;
2016  
2017      if (new_bs) {
2018          QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2019          if (new_bs->quiesce_counter && child->role->drained_begin) {
2020              for (i = 0; i < new_bs->quiesce_counter; i++) {
2021                  child->role->drained_begin(child);
2022              }
2023          }
2024  
2025          /* Attach only after starting new drained sections, so that recursive
2026           * drain sections coming from @child don't get an extra .drained_begin
2027           * callback. */
2028          if (child->role->attach) {
2029              child->role->attach(child);
2030          }
2031      }
2032  }
2033  
2034  /*
2035   * Updates @child to change its reference to point to @new_bs, including
2036   * checking and applying the necessary permisson updates both to the old node
2037   * and to @new_bs.
2038   *
2039   * NULL is passed as @new_bs for removing the reference before freeing @child.
2040   *
2041   * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2042   * function uses bdrv_set_perm() to update the permissions according to the new
2043   * reference that @new_bs gets.
2044   */
2045  static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2046  {
2047      BlockDriverState *old_bs = child->bs;
2048      uint64_t perm, shared_perm;
2049  
2050      bdrv_replace_child_noperm(child, new_bs);
2051  
2052      if (old_bs) {
2053          /* Update permissions for old node. This is guaranteed to succeed
2054           * because we're just taking a parent away, so we're loosening
2055           * restrictions. */
2056          bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2057          bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
2058          bdrv_set_perm(old_bs, perm, shared_perm);
2059      }
2060  
2061      if (new_bs) {
2062          bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2063          bdrv_set_perm(new_bs, perm, shared_perm);
2064      }
2065  }
2066  
2067  BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2068                                    const char *child_name,
2069                                    const BdrvChildRole *child_role,
2070                                    uint64_t perm, uint64_t shared_perm,
2071                                    void *opaque, Error **errp)
2072  {
2073      BdrvChild *child;
2074      int ret;
2075  
2076      ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2077      if (ret < 0) {
2078          bdrv_abort_perm_update(child_bs);
2079          return NULL;
2080      }
2081  
2082      child = g_new(BdrvChild, 1);
2083      *child = (BdrvChild) {
2084          .bs             = NULL,
2085          .name           = g_strdup(child_name),
2086          .role           = child_role,
2087          .perm           = perm,
2088          .shared_perm    = shared_perm,
2089          .opaque         = opaque,
2090      };
2091  
2092      /* This performs the matching bdrv_set_perm() for the above check. */
2093      bdrv_replace_child(child, child_bs);
2094  
2095      return child;
2096  }
2097  
2098  BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2099                               BlockDriverState *child_bs,
2100                               const char *child_name,
2101                               const BdrvChildRole *child_role,
2102                               Error **errp)
2103  {
2104      BdrvChild *child;
2105      uint64_t perm, shared_perm;
2106  
2107      bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2108  
2109      assert(parent_bs->drv);
2110      assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
2111      bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2112                      perm, shared_perm, &perm, &shared_perm);
2113  
2114      child = bdrv_root_attach_child(child_bs, child_name, child_role,
2115                                     perm, shared_perm, parent_bs, errp);
2116      if (child == NULL) {
2117          return NULL;
2118      }
2119  
2120      QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2121      return child;
2122  }
2123  
2124  static void bdrv_detach_child(BdrvChild *child)
2125  {
2126      if (child->next.le_prev) {
2127          QLIST_REMOVE(child, next);
2128          child->next.le_prev = NULL;
2129      }
2130  
2131      bdrv_replace_child(child, NULL);
2132  
2133      g_free(child->name);
2134      g_free(child);
2135  }
2136  
2137  void bdrv_root_unref_child(BdrvChild *child)
2138  {
2139      BlockDriverState *child_bs;
2140  
2141      child_bs = child->bs;
2142      bdrv_detach_child(child);
2143      bdrv_unref(child_bs);
2144  }
2145  
2146  void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2147  {
2148      if (child == NULL) {
2149          return;
2150      }
2151  
2152      if (child->bs->inherits_from == parent) {
2153          BdrvChild *c;
2154  
2155          /* Remove inherits_from only when the last reference between parent and
2156           * child->bs goes away. */
2157          QLIST_FOREACH(c, &parent->children, next) {
2158              if (c != child && c->bs == child->bs) {
2159                  break;
2160              }
2161          }
2162          if (c == NULL) {
2163              child->bs->inherits_from = NULL;
2164          }
2165      }
2166  
2167      bdrv_root_unref_child(child);
2168  }
2169  
2170  
2171  static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2172  {
2173      BdrvChild *c;
2174      QLIST_FOREACH(c, &bs->parents, next_parent) {
2175          if (c->role->change_media) {
2176              c->role->change_media(c, load);
2177          }
2178      }
2179  }
2180  
2181  static void bdrv_parent_cb_resize(BlockDriverState *bs)
2182  {
2183      BdrvChild *c;
2184      QLIST_FOREACH(c, &bs->parents, next_parent) {
2185          if (c->role->resize) {
2186              c->role->resize(c);
2187          }
2188      }
2189  }
2190  
2191  /*
2192   * Sets the backing file link of a BDS. A new reference is created; callers
2193   * which don't need their own reference any more must call bdrv_unref().
2194   */
2195  void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2196                           Error **errp)
2197  {
2198      if (backing_hd) {
2199          bdrv_ref(backing_hd);
2200      }
2201  
2202      if (bs->backing) {
2203          bdrv_unref_child(bs, bs->backing);
2204      }
2205  
2206      if (!backing_hd) {
2207          bs->backing = NULL;
2208          goto out;
2209      }
2210  
2211      bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2212                                      errp);
2213      if (!bs->backing) {
2214          bdrv_unref(backing_hd);
2215      }
2216  
2217      bdrv_refresh_filename(bs);
2218  
2219  out:
2220      bdrv_refresh_limits(bs, NULL);
2221  }
2222  
2223  /*
2224   * Opens the backing file for a BlockDriverState if not yet open
2225   *
2226   * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2227   * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2228   * itself, all options starting with "${bdref_key}." are considered part of the
2229   * BlockdevRef.
2230   *
2231   * TODO Can this be unified with bdrv_open_image()?
2232   */
2233  int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2234                             const char *bdref_key, Error **errp)
2235  {
2236      char *backing_filename = g_malloc0(PATH_MAX);
2237      char *bdref_key_dot;
2238      const char *reference = NULL;
2239      int ret = 0;
2240      BlockDriverState *backing_hd;
2241      QDict *options;
2242      QDict *tmp_parent_options = NULL;
2243      Error *local_err = NULL;
2244  
2245      if (bs->backing != NULL) {
2246          goto free_exit;
2247      }
2248  
2249      /* NULL means an empty set of options */
2250      if (parent_options == NULL) {
2251          tmp_parent_options = qdict_new();
2252          parent_options = tmp_parent_options;
2253      }
2254  
2255      bs->open_flags &= ~BDRV_O_NO_BACKING;
2256  
2257      bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2258      qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2259      g_free(bdref_key_dot);
2260  
2261      /*
2262       * Caution: while qdict_get_try_str() is fine, getting non-string
2263       * types would require more care.  When @parent_options come from
2264       * -blockdev or blockdev_add, its members are typed according to
2265       * the QAPI schema, but when they come from -drive, they're all
2266       * QString.
2267       */
2268      reference = qdict_get_try_str(parent_options, bdref_key);
2269      if (reference || qdict_haskey(options, "file.filename")) {
2270          backing_filename[0] = '\0';
2271      } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2272          QDECREF(options);
2273          goto free_exit;
2274      } else {
2275          bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2276                                         &local_err);
2277          if (local_err) {
2278              ret = -EINVAL;
2279              error_propagate(errp, local_err);
2280              QDECREF(options);
2281              goto free_exit;
2282          }
2283      }
2284  
2285      if (!bs->drv || !bs->drv->supports_backing) {
2286          ret = -EINVAL;
2287          error_setg(errp, "Driver doesn't support backing files");
2288          QDECREF(options);
2289          goto free_exit;
2290      }
2291  
2292      if (!reference &&
2293          bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2294          qdict_put_str(options, "driver", bs->backing_format);
2295      }
2296  
2297      backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2298                                     reference, options, 0, bs, &child_backing,
2299                                     errp);
2300      if (!backing_hd) {
2301          bs->open_flags |= BDRV_O_NO_BACKING;
2302          error_prepend(errp, "Could not open backing file: ");
2303          ret = -EINVAL;
2304          goto free_exit;
2305      }
2306      bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
2307  
2308      /* Hook up the backing file link; drop our reference, bs owns the
2309       * backing_hd reference now */
2310      bdrv_set_backing_hd(bs, backing_hd, &local_err);
2311      bdrv_unref(backing_hd);
2312      if (local_err) {
2313          error_propagate(errp, local_err);
2314          ret = -EINVAL;
2315          goto free_exit;
2316      }
2317  
2318      qdict_del(parent_options, bdref_key);
2319  
2320  free_exit:
2321      g_free(backing_filename);
2322      QDECREF(tmp_parent_options);
2323      return ret;
2324  }
2325  
2326  static BlockDriverState *
2327  bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2328                     BlockDriverState *parent, const BdrvChildRole *child_role,
2329                     bool allow_none, Error **errp)
2330  {
2331      BlockDriverState *bs = NULL;
2332      QDict *image_options;
2333      char *bdref_key_dot;
2334      const char *reference;
2335  
2336      assert(child_role != NULL);
2337  
2338      bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2339      qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2340      g_free(bdref_key_dot);
2341  
2342      /*
2343       * Caution: while qdict_get_try_str() is fine, getting non-string
2344       * types would require more care.  When @options come from
2345       * -blockdev or blockdev_add, its members are typed according to
2346       * the QAPI schema, but when they come from -drive, they're all
2347       * QString.
2348       */
2349      reference = qdict_get_try_str(options, bdref_key);
2350      if (!filename && !reference && !qdict_size(image_options)) {
2351          if (!allow_none) {
2352              error_setg(errp, "A block device must be specified for \"%s\"",
2353                         bdref_key);
2354          }
2355          QDECREF(image_options);
2356          goto done;
2357      }
2358  
2359      bs = bdrv_open_inherit(filename, reference, image_options, 0,
2360                             parent, child_role, errp);
2361      if (!bs) {
2362          goto done;
2363      }
2364  
2365  done:
2366      qdict_del(options, bdref_key);
2367      return bs;
2368  }
2369  
2370  /*
2371   * Opens a disk image whose options are given as BlockdevRef in another block
2372   * device's options.
2373   *
2374   * If allow_none is true, no image will be opened if filename is false and no
2375   * BlockdevRef is given. NULL will be returned, but errp remains unset.
2376   *
2377   * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2378   * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2379   * itself, all options starting with "${bdref_key}." are considered part of the
2380   * BlockdevRef.
2381   *
2382   * The BlockdevRef will be removed from the options QDict.
2383   */
2384  BdrvChild *bdrv_open_child(const char *filename,
2385                             QDict *options, const char *bdref_key,
2386                             BlockDriverState *parent,
2387                             const BdrvChildRole *child_role,
2388                             bool allow_none, Error **errp)
2389  {
2390      BdrvChild *c;
2391      BlockDriverState *bs;
2392  
2393      bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2394                              allow_none, errp);
2395      if (bs == NULL) {
2396          return NULL;
2397      }
2398  
2399      c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2400      if (!c) {
2401          bdrv_unref(bs);
2402          return NULL;
2403      }
2404  
2405      return c;
2406  }
2407  
2408  static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2409                                                     int flags,
2410                                                     QDict *snapshot_options,
2411                                                     Error **errp)
2412  {
2413      /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2414      char *tmp_filename = g_malloc0(PATH_MAX + 1);
2415      int64_t total_size;
2416      QemuOpts *opts = NULL;
2417      BlockDriverState *bs_snapshot = NULL;
2418      Error *local_err = NULL;
2419      int ret;
2420  
2421      /* if snapshot, we create a temporary backing file and open it
2422         instead of opening 'filename' directly */
2423  
2424      /* Get the required size from the image */
2425      total_size = bdrv_getlength(bs);
2426      if (total_size < 0) {
2427          error_setg_errno(errp, -total_size, "Could not get image size");
2428          goto out;
2429      }
2430  
2431      /* Create the temporary image */
2432      ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2433      if (ret < 0) {
2434          error_setg_errno(errp, -ret, "Could not get temporary filename");
2435          goto out;
2436      }
2437  
2438      opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2439                              &error_abort);
2440      qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2441      ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2442      qemu_opts_del(opts);
2443      if (ret < 0) {
2444          error_prepend(errp, "Could not create temporary overlay '%s': ",
2445                        tmp_filename);
2446          goto out;
2447      }
2448  
2449      /* Prepare options QDict for the temporary file */
2450      qdict_put_str(snapshot_options, "file.driver", "file");
2451      qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2452      qdict_put_str(snapshot_options, "driver", "qcow2");
2453  
2454      bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2455      snapshot_options = NULL;
2456      if (!bs_snapshot) {
2457          goto out;
2458      }
2459  
2460      /* bdrv_append() consumes a strong reference to bs_snapshot
2461       * (i.e. it will call bdrv_unref() on it) even on error, so in
2462       * order to be able to return one, we have to increase
2463       * bs_snapshot's refcount here */
2464      bdrv_ref(bs_snapshot);
2465      bdrv_append(bs_snapshot, bs, &local_err);
2466      if (local_err) {
2467          error_propagate(errp, local_err);
2468          bs_snapshot = NULL;
2469          goto out;
2470      }
2471  
2472  out:
2473      QDECREF(snapshot_options);
2474      g_free(tmp_filename);
2475      return bs_snapshot;
2476  }
2477  
2478  /*
2479   * Opens a disk image (raw, qcow2, vmdk, ...)
2480   *
2481   * options is a QDict of options to pass to the block drivers, or NULL for an
2482   * empty set of options. The reference to the QDict belongs to the block layer
2483   * after the call (even on failure), so if the caller intends to reuse the
2484   * dictionary, it needs to use QINCREF() before calling bdrv_open.
2485   *
2486   * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2487   * If it is not NULL, the referenced BDS will be reused.
2488   *
2489   * The reference parameter may be used to specify an existing block device which
2490   * should be opened. If specified, neither options nor a filename may be given,
2491   * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2492   */
2493  static BlockDriverState *bdrv_open_inherit(const char *filename,
2494                                             const char *reference,
2495                                             QDict *options, int flags,
2496                                             BlockDriverState *parent,
2497                                             const BdrvChildRole *child_role,
2498                                             Error **errp)
2499  {
2500      int ret;
2501      BlockBackend *file = NULL;
2502      BlockDriverState *bs;
2503      BlockDriver *drv = NULL;
2504      const char *drvname;
2505      const char *backing;
2506      Error *local_err = NULL;
2507      QDict *snapshot_options = NULL;
2508      int snapshot_flags = 0;
2509  
2510      assert(!child_role || !flags);
2511      assert(!child_role == !parent);
2512  
2513      if (reference) {
2514          bool options_non_empty = options ? qdict_size(options) : false;
2515          QDECREF(options);
2516  
2517          if (filename || options_non_empty) {
2518              error_setg(errp, "Cannot reference an existing block device with "
2519                         "additional options or a new filename");
2520              return NULL;
2521          }
2522  
2523          bs = bdrv_lookup_bs(reference, reference, errp);
2524          if (!bs) {
2525              return NULL;
2526          }
2527  
2528          bdrv_ref(bs);
2529          return bs;
2530      }
2531  
2532      bs = bdrv_new();
2533  
2534      /* NULL means an empty set of options */
2535      if (options == NULL) {
2536          options = qdict_new();
2537      }
2538  
2539      /* json: syntax counts as explicit options, as if in the QDict */
2540      parse_json_protocol(options, &filename, &local_err);
2541      if (local_err) {
2542          goto fail;
2543      }
2544  
2545      bs->explicit_options = qdict_clone_shallow(options);
2546  
2547      if (child_role) {
2548          bs->inherits_from = parent;
2549          child_role->inherit_options(&flags, options,
2550                                      parent->open_flags, parent->options);
2551      }
2552  
2553      ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2554      if (local_err) {
2555          goto fail;
2556      }
2557  
2558      /*
2559       * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2560       * Caution: getting a boolean member of @options requires care.
2561       * When @options come from -blockdev or blockdev_add, members are
2562       * typed according to the QAPI schema, but when they come from
2563       * -drive, they're all QString.
2564       */
2565      if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2566          !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2567          flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2568      } else {
2569          flags &= ~BDRV_O_RDWR;
2570      }
2571  
2572      if (flags & BDRV_O_SNAPSHOT) {
2573          snapshot_options = qdict_new();
2574          bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2575                                     flags, options);
2576          /* Let bdrv_backing_options() override "read-only" */
2577          qdict_del(options, BDRV_OPT_READ_ONLY);
2578          bdrv_backing_options(&flags, options, flags, options);
2579      }
2580  
2581      bs->open_flags = flags;
2582      bs->options = options;
2583      options = qdict_clone_shallow(options);
2584  
2585      /* Find the right image format driver */
2586      /* See cautionary note on accessing @options above */
2587      drvname = qdict_get_try_str(options, "driver");
2588      if (drvname) {
2589          drv = bdrv_find_format(drvname);
2590          if (!drv) {
2591              error_setg(errp, "Unknown driver: '%s'", drvname);
2592              goto fail;
2593          }
2594      }
2595  
2596      assert(drvname || !(flags & BDRV_O_PROTOCOL));
2597  
2598      /* See cautionary note on accessing @options above */
2599      backing = qdict_get_try_str(options, "backing");
2600      if (backing && *backing == '\0') {
2601          flags |= BDRV_O_NO_BACKING;
2602          qdict_del(options, "backing");
2603      }
2604  
2605      /* Open image file without format layer. This BlockBackend is only used for
2606       * probing, the block drivers will do their own bdrv_open_child() for the
2607       * same BDS, which is why we put the node name back into options. */
2608      if ((flags & BDRV_O_PROTOCOL) == 0) {
2609          BlockDriverState *file_bs;
2610  
2611          file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2612                                       &child_file, true, &local_err);
2613          if (local_err) {
2614              goto fail;
2615          }
2616          if (file_bs != NULL) {
2617              /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2618               * looking at the header to guess the image format. This works even
2619               * in cases where a guest would not see a consistent state. */
2620              file = blk_new(0, BLK_PERM_ALL);
2621              blk_insert_bs(file, file_bs, &local_err);
2622              bdrv_unref(file_bs);
2623              if (local_err) {
2624                  goto fail;
2625              }
2626  
2627              qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2628          }
2629      }
2630  
2631      /* Image format probing */
2632      bs->probed = !drv;
2633      if (!drv && file) {
2634          ret = find_image_format(file, filename, &drv, &local_err);
2635          if (ret < 0) {
2636              goto fail;
2637          }
2638          /*
2639           * This option update would logically belong in bdrv_fill_options(),
2640           * but we first need to open bs->file for the probing to work, while
2641           * opening bs->file already requires the (mostly) final set of options
2642           * so that cache mode etc. can be inherited.
2643           *
2644           * Adding the driver later is somewhat ugly, but it's not an option
2645           * that would ever be inherited, so it's correct. We just need to make
2646           * sure to update both bs->options (which has the full effective
2647           * options for bs) and options (which has file.* already removed).
2648           */
2649          qdict_put_str(bs->options, "driver", drv->format_name);
2650          qdict_put_str(options, "driver", drv->format_name);
2651      } else if (!drv) {
2652          error_setg(errp, "Must specify either driver or file");
2653          goto fail;
2654      }
2655  
2656      /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2657      assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2658      /* file must be NULL if a protocol BDS is about to be created
2659       * (the inverse results in an error message from bdrv_open_common()) */
2660      assert(!(flags & BDRV_O_PROTOCOL) || !file);
2661  
2662      /* Open the image */
2663      ret = bdrv_open_common(bs, file, options, &local_err);
2664      if (ret < 0) {
2665          goto fail;
2666      }
2667  
2668      if (file) {
2669          blk_unref(file);
2670          file = NULL;
2671      }
2672  
2673      /* If there is a backing file, use it */
2674      if ((flags & BDRV_O_NO_BACKING) == 0) {
2675          ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2676          if (ret < 0) {
2677              goto close_and_fail;
2678          }
2679      }
2680  
2681      bdrv_refresh_filename(bs);
2682  
2683      /* Check if any unknown options were used */
2684      if (qdict_size(options) != 0) {
2685          const QDictEntry *entry = qdict_first(options);
2686          if (flags & BDRV_O_PROTOCOL) {
2687              error_setg(errp, "Block protocol '%s' doesn't support the option "
2688                         "'%s'", drv->format_name, entry->key);
2689          } else {
2690              error_setg(errp,
2691                         "Block format '%s' does not support the option '%s'",
2692                         drv->format_name, entry->key);
2693          }
2694  
2695          goto close_and_fail;
2696      }
2697  
2698      bdrv_parent_cb_change_media(bs, true);
2699  
2700      QDECREF(options);
2701  
2702      /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2703       * temporary snapshot afterwards. */
2704      if (snapshot_flags) {
2705          BlockDriverState *snapshot_bs;
2706          snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2707                                                  snapshot_options, &local_err);
2708          snapshot_options = NULL;
2709          if (local_err) {
2710              goto close_and_fail;
2711          }
2712          /* We are not going to return bs but the overlay on top of it
2713           * (snapshot_bs); thus, we have to drop the strong reference to bs
2714           * (which we obtained by calling bdrv_new()). bs will not be deleted,
2715           * though, because the overlay still has a reference to it. */
2716          bdrv_unref(bs);
2717          bs = snapshot_bs;
2718      }
2719  
2720      return bs;
2721  
2722  fail:
2723      blk_unref(file);
2724      QDECREF(snapshot_options);
2725      QDECREF(bs->explicit_options);
2726      QDECREF(bs->options);
2727      QDECREF(options);
2728      bs->options = NULL;
2729      bs->explicit_options = NULL;
2730      bdrv_unref(bs);
2731      error_propagate(errp, local_err);
2732      return NULL;
2733  
2734  close_and_fail:
2735      bdrv_unref(bs);
2736      QDECREF(snapshot_options);
2737      QDECREF(options);
2738      error_propagate(errp, local_err);
2739      return NULL;
2740  }
2741  
2742  BlockDriverState *bdrv_open(const char *filename, const char *reference,
2743                              QDict *options, int flags, Error **errp)
2744  {
2745      return bdrv_open_inherit(filename, reference, options, flags, NULL,
2746                               NULL, errp);
2747  }
2748  
2749  /*
2750   * Adds a BlockDriverState to a simple queue for an atomic, transactional
2751   * reopen of multiple devices.
2752   *
2753   * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2754   * already performed, or alternatively may be NULL a new BlockReopenQueue will
2755   * be created and initialized. This newly created BlockReopenQueue should be
2756   * passed back in for subsequent calls that are intended to be of the same
2757   * atomic 'set'.
2758   *
2759   * bs is the BlockDriverState to add to the reopen queue.
2760   *
2761   * options contains the changed options for the associated bs
2762   * (the BlockReopenQueue takes ownership)
2763   *
2764   * flags contains the open flags for the associated bs
2765   *
2766   * returns a pointer to bs_queue, which is either the newly allocated
2767   * bs_queue, or the existing bs_queue being used.
2768   *
2769   * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2770   */
2771  static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2772                                                   BlockDriverState *bs,
2773                                                   QDict *options,
2774                                                   int flags,
2775                                                   const BdrvChildRole *role,
2776                                                   QDict *parent_options,
2777                                                   int parent_flags)
2778  {
2779      assert(bs != NULL);
2780  
2781      BlockReopenQueueEntry *bs_entry;
2782      BdrvChild *child;
2783      QDict *old_options, *explicit_options;
2784  
2785      /* Make sure that the caller remembered to use a drained section. This is
2786       * important to avoid graph changes between the recursive queuing here and
2787       * bdrv_reopen_multiple(). */
2788      assert(bs->quiesce_counter > 0);
2789  
2790      if (bs_queue == NULL) {
2791          bs_queue = g_new0(BlockReopenQueue, 1);
2792          QSIMPLEQ_INIT(bs_queue);
2793      }
2794  
2795      if (!options) {
2796          options = qdict_new();
2797      }
2798  
2799      /* Check if this BlockDriverState is already in the queue */
2800      QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2801          if (bs == bs_entry->state.bs) {
2802              break;
2803          }
2804      }
2805  
2806      /*
2807       * Precedence of options:
2808       * 1. Explicitly passed in options (highest)
2809       * 2. Set in flags (only for top level)
2810       * 3. Retained from explicitly set options of bs
2811       * 4. Inherited from parent node
2812       * 5. Retained from effective options of bs
2813       */
2814  
2815      if (!parent_options) {
2816          /*
2817           * Any setting represented by flags is always updated. If the
2818           * corresponding QDict option is set, it takes precedence. Otherwise
2819           * the flag is translated into a QDict option. The old setting of bs is
2820           * not considered.
2821           */
2822          update_options_from_flags(options, flags);
2823      }
2824  
2825      /* Old explicitly set values (don't overwrite by inherited value) */
2826      if (bs_entry) {
2827          old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2828      } else {
2829          old_options = qdict_clone_shallow(bs->explicit_options);
2830      }
2831      bdrv_join_options(bs, options, old_options);
2832      QDECREF(old_options);
2833  
2834      explicit_options = qdict_clone_shallow(options);
2835  
2836      /* Inherit from parent node */
2837      if (parent_options) {
2838          assert(!flags);
2839          role->inherit_options(&flags, options, parent_flags, parent_options);
2840      }
2841  
2842      /* Old values are used for options that aren't set yet */
2843      old_options = qdict_clone_shallow(bs->options);
2844      bdrv_join_options(bs, options, old_options);
2845      QDECREF(old_options);
2846  
2847      /* bdrv_open_inherit() sets and clears some additional flags internally */
2848      flags &= ~BDRV_O_PROTOCOL;
2849      if (flags & BDRV_O_RDWR) {
2850          flags |= BDRV_O_ALLOW_RDWR;
2851      }
2852  
2853      if (!bs_entry) {
2854          bs_entry = g_new0(BlockReopenQueueEntry, 1);
2855          QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2856      } else {
2857          QDECREF(bs_entry->state.options);
2858          QDECREF(bs_entry->state.explicit_options);
2859      }
2860  
2861      bs_entry->state.bs = bs;
2862      bs_entry->state.options = options;
2863      bs_entry->state.explicit_options = explicit_options;
2864      bs_entry->state.flags = flags;
2865  
2866      /* This needs to be overwritten in bdrv_reopen_prepare() */
2867      bs_entry->state.perm = UINT64_MAX;
2868      bs_entry->state.shared_perm = 0;
2869  
2870      QLIST_FOREACH(child, &bs->children, next) {
2871          QDict *new_child_options;
2872          char *child_key_dot;
2873  
2874          /* reopen can only change the options of block devices that were
2875           * implicitly created and inherited options. For other (referenced)
2876           * block devices, a syntax like "backing.foo" results in an error. */
2877          if (child->bs->inherits_from != bs) {
2878              continue;
2879          }
2880  
2881          child_key_dot = g_strdup_printf("%s.", child->name);
2882          qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2883          g_free(child_key_dot);
2884  
2885          bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2886                                  child->role, options, flags);
2887      }
2888  
2889      return bs_queue;
2890  }
2891  
2892  BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2893                                      BlockDriverState *bs,
2894                                      QDict *options, int flags)
2895  {
2896      return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2897                                     NULL, NULL, 0);
2898  }
2899  
2900  /*
2901   * Reopen multiple BlockDriverStates atomically & transactionally.
2902   *
2903   * The queue passed in (bs_queue) must have been built up previous
2904   * via bdrv_reopen_queue().
2905   *
2906   * Reopens all BDS specified in the queue, with the appropriate
2907   * flags.  All devices are prepared for reopen, and failure of any
2908   * device will cause all device changes to be abandonded, and intermediate
2909   * data cleaned up.
2910   *
2911   * If all devices prepare successfully, then the changes are committed
2912   * to all devices.
2913   *
2914   * All affected nodes must be drained between bdrv_reopen_queue() and
2915   * bdrv_reopen_multiple().
2916   */
2917  int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2918  {
2919      int ret = -1;
2920      BlockReopenQueueEntry *bs_entry, *next;
2921      Error *local_err = NULL;
2922  
2923      assert(bs_queue != NULL);
2924  
2925      QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2926          assert(bs_entry->state.bs->quiesce_counter > 0);
2927          if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2928              error_propagate(errp, local_err);
2929              goto cleanup;
2930          }
2931          bs_entry->prepared = true;
2932      }
2933  
2934      /* If we reach this point, we have success and just need to apply the
2935       * changes
2936       */
2937      QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2938          bdrv_reopen_commit(&bs_entry->state);
2939      }
2940  
2941      ret = 0;
2942  
2943  cleanup:
2944      QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
2945          if (ret && bs_entry->prepared) {
2946              bdrv_reopen_abort(&bs_entry->state);
2947          } else if (ret) {
2948              QDECREF(bs_entry->state.explicit_options);
2949          }
2950          QDECREF(bs_entry->state.options);
2951          g_free(bs_entry);
2952      }
2953      g_free(bs_queue);
2954  
2955      return ret;
2956  }
2957  
2958  
2959  /* Reopen a single BlockDriverState with the specified flags. */
2960  int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
2961  {
2962      int ret = -1;
2963      Error *local_err = NULL;
2964      BlockReopenQueue *queue;
2965  
2966      bdrv_subtree_drained_begin(bs);
2967  
2968      queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
2969      ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
2970      if (local_err != NULL) {
2971          error_propagate(errp, local_err);
2972      }
2973  
2974      bdrv_subtree_drained_end(bs);
2975  
2976      return ret;
2977  }
2978  
2979  static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
2980                                                            BdrvChild *c)
2981  {
2982      BlockReopenQueueEntry *entry;
2983  
2984      QSIMPLEQ_FOREACH(entry, q, entry) {
2985          BlockDriverState *bs = entry->state.bs;
2986          BdrvChild *child;
2987  
2988          QLIST_FOREACH(child, &bs->children, next) {
2989              if (child == c) {
2990                  return entry;
2991              }
2992          }
2993      }
2994  
2995      return NULL;
2996  }
2997  
2998  static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
2999                               uint64_t *perm, uint64_t *shared)
3000  {
3001      BdrvChild *c;
3002      BlockReopenQueueEntry *parent;
3003      uint64_t cumulative_perms = 0;
3004      uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3005  
3006      QLIST_FOREACH(c, &bs->parents, next_parent) {
3007          parent = find_parent_in_reopen_queue(q, c);
3008          if (!parent) {
3009              cumulative_perms |= c->perm;
3010              cumulative_shared_perms &= c->shared_perm;
3011          } else {
3012              uint64_t nperm, nshared;
3013  
3014              bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3015                              parent->state.perm, parent->state.shared_perm,
3016                              &nperm, &nshared);
3017  
3018              cumulative_perms |= nperm;
3019              cumulative_shared_perms &= nshared;
3020          }
3021      }
3022      *perm = cumulative_perms;
3023      *shared = cumulative_shared_perms;
3024  }
3025  
3026  /*
3027   * Prepares a BlockDriverState for reopen. All changes are staged in the
3028   * 'opaque' field of the BDRVReopenState, which is used and allocated by
3029   * the block driver layer .bdrv_reopen_prepare()
3030   *
3031   * bs is the BlockDriverState to reopen
3032   * flags are the new open flags
3033   * queue is the reopen queue
3034   *
3035   * Returns 0 on success, non-zero on error.  On error errp will be set
3036   * as well.
3037   *
3038   * On failure, bdrv_reopen_abort() will be called to clean up any data.
3039   * It is the responsibility of the caller to then call the abort() or
3040   * commit() for any other BDS that have been left in a prepare() state
3041   *
3042   */
3043  int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3044                          Error **errp)
3045  {
3046      int ret = -1;
3047      Error *local_err = NULL;
3048      BlockDriver *drv;
3049      QemuOpts *opts;
3050      const char *value;
3051      bool read_only;
3052  
3053      assert(reopen_state != NULL);
3054      assert(reopen_state->bs->drv != NULL);
3055      drv = reopen_state->bs->drv;
3056  
3057      /* Process generic block layer options */
3058      opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3059      qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3060      if (local_err) {
3061          error_propagate(errp, local_err);
3062          ret = -EINVAL;
3063          goto error;
3064      }
3065  
3066      update_flags_from_options(&reopen_state->flags, opts);
3067  
3068      /* node-name and driver must be unchanged. Put them back into the QDict, so
3069       * that they are checked at the end of this function. */
3070      value = qemu_opt_get(opts, "node-name");
3071      if (value) {
3072          qdict_put_str(reopen_state->options, "node-name", value);
3073      }
3074  
3075      value = qemu_opt_get(opts, "driver");
3076      if (value) {
3077          qdict_put_str(reopen_state->options, "driver", value);
3078      }
3079  
3080      /* If we are to stay read-only, do not allow permission change
3081       * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3082       * not set, or if the BDS still has copy_on_read enabled */
3083      read_only = !(reopen_state->flags & BDRV_O_RDWR);
3084      ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3085      if (local_err) {
3086          error_propagate(errp, local_err);
3087          goto error;
3088      }
3089  
3090      /* Calculate required permissions after reopening */
3091      bdrv_reopen_perm(queue, reopen_state->bs,
3092                       &reopen_state->perm, &reopen_state->shared_perm);
3093  
3094      ret = bdrv_flush(reopen_state->bs);
3095      if (ret) {
3096          error_setg_errno(errp, -ret, "Error flushing drive");
3097          goto error;
3098      }
3099  
3100      if (drv->bdrv_reopen_prepare) {
3101          ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3102          if (ret) {
3103              if (local_err != NULL) {
3104                  error_propagate(errp, local_err);
3105              } else {
3106                  error_setg(errp, "failed while preparing to reopen image '%s'",
3107                             reopen_state->bs->filename);
3108              }
3109              goto error;
3110          }
3111      } else {
3112          /* It is currently mandatory to have a bdrv_reopen_prepare()
3113           * handler for each supported drv. */
3114          error_setg(errp, "Block format '%s' used by node '%s' "
3115                     "does not support reopening files", drv->format_name,
3116                     bdrv_get_device_or_node_name(reopen_state->bs));
3117          ret = -1;
3118          goto error;
3119      }
3120  
3121      /* Options that are not handled are only okay if they are unchanged
3122       * compared to the old state. It is expected that some options are only
3123       * used for the initial open, but not reopen (e.g. filename) */
3124      if (qdict_size(reopen_state->options)) {
3125          const QDictEntry *entry = qdict_first(reopen_state->options);
3126  
3127          do {
3128              QObject *new = entry->value;
3129              QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3130  
3131              /*
3132               * TODO: When using -drive to specify blockdev options, all values
3133               * will be strings; however, when using -blockdev, blockdev-add or
3134               * filenames using the json:{} pseudo-protocol, they will be
3135               * correctly typed.
3136               * In contrast, reopening options are (currently) always strings
3137               * (because you can only specify them through qemu-io; all other
3138               * callers do not specify any options).
3139               * Therefore, when using anything other than -drive to create a BDS,
3140               * this cannot detect non-string options as unchanged, because
3141               * qobject_is_equal() always returns false for objects of different
3142               * type.  In the future, this should be remedied by correctly typing
3143               * all options.  For now, this is not too big of an issue because
3144               * the user can simply omit options which cannot be changed anyway,
3145               * so they will stay unchanged.
3146               */
3147              if (!qobject_is_equal(new, old)) {
3148                  error_setg(errp, "Cannot change the option '%s'", entry->key);
3149                  ret = -EINVAL;
3150                  goto error;
3151              }
3152          } while ((entry = qdict_next(reopen_state->options, entry)));
3153      }
3154  
3155      ret = bdrv_check_perm(reopen_state->bs, queue, reopen_state->perm,
3156                            reopen_state->shared_perm, NULL, errp);
3157      if (ret < 0) {
3158          goto error;
3159      }
3160  
3161      ret = 0;
3162  
3163  error:
3164      qemu_opts_del(opts);
3165      return ret;
3166  }
3167  
3168  /*
3169   * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3170   * makes them final by swapping the staging BlockDriverState contents into
3171   * the active BlockDriverState contents.
3172   */
3173  void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3174  {
3175      BlockDriver *drv;
3176      BlockDriverState *bs;
3177      bool old_can_write, new_can_write;
3178  
3179      assert(reopen_state != NULL);
3180      bs = reopen_state->bs;
3181      drv = bs->drv;
3182      assert(drv != NULL);
3183  
3184      old_can_write =
3185          !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3186  
3187      /* If there are any driver level actions to take */
3188      if (drv->bdrv_reopen_commit) {
3189          drv->bdrv_reopen_commit(reopen_state);
3190      }
3191  
3192      /* set BDS specific flags now */
3193      QDECREF(bs->explicit_options);
3194  
3195      bs->explicit_options   = reopen_state->explicit_options;
3196      bs->open_flags         = reopen_state->flags;
3197      bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3198  
3199      bdrv_refresh_limits(bs, NULL);
3200  
3201      bdrv_set_perm(reopen_state->bs, reopen_state->perm,
3202                    reopen_state->shared_perm);
3203  
3204      new_can_write =
3205          !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3206      if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3207          Error *local_err = NULL;
3208          if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3209              /* This is not fatal, bitmaps just left read-only, so all following
3210               * writes will fail. User can remove read-only bitmaps to unblock
3211               * writes.
3212               */
3213              error_reportf_err(local_err,
3214                                "%s: Failed to make dirty bitmaps writable: ",
3215                                bdrv_get_node_name(bs));
3216          }
3217      }
3218  }
3219  
3220  /*
3221   * Abort the reopen, and delete and free the staged changes in
3222   * reopen_state
3223   */
3224  void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3225  {
3226      BlockDriver *drv;
3227  
3228      assert(reopen_state != NULL);
3229      drv = reopen_state->bs->drv;
3230      assert(drv != NULL);
3231  
3232      if (drv->bdrv_reopen_abort) {
3233          drv->bdrv_reopen_abort(reopen_state);
3234      }
3235  
3236      QDECREF(reopen_state->explicit_options);
3237  
3238      bdrv_abort_perm_update(reopen_state->bs);
3239  }
3240  
3241  
3242  static void bdrv_close(BlockDriverState *bs)
3243  {
3244      BdrvAioNotifier *ban, *ban_next;
3245      BdrvChild *child, *next;
3246  
3247      assert(!bs->job);
3248      assert(!bs->refcnt);
3249  
3250      bdrv_drained_begin(bs); /* complete I/O */
3251      bdrv_flush(bs);
3252      bdrv_drain(bs); /* in case flush left pending I/O */
3253  
3254      if (bs->drv) {
3255          bs->drv->bdrv_close(bs);
3256          bs->drv = NULL;
3257      }
3258  
3259      bdrv_set_backing_hd(bs, NULL, &error_abort);
3260  
3261      if (bs->file != NULL) {
3262          bdrv_unref_child(bs, bs->file);
3263          bs->file = NULL;
3264      }
3265  
3266      QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3267          /* TODO Remove bdrv_unref() from drivers' close function and use
3268           * bdrv_unref_child() here */
3269          if (child->bs->inherits_from == bs) {
3270              child->bs->inherits_from = NULL;
3271          }
3272          bdrv_detach_child(child);
3273      }
3274  
3275      g_free(bs->opaque);
3276      bs->opaque = NULL;
3277      atomic_set(&bs->copy_on_read, 0);
3278      bs->backing_file[0] = '\0';
3279      bs->backing_format[0] = '\0';
3280      bs->total_sectors = 0;
3281      bs->encrypted = false;
3282      bs->sg = false;
3283      QDECREF(bs->options);
3284      QDECREF(bs->explicit_options);
3285      bs->options = NULL;
3286      bs->explicit_options = NULL;
3287      QDECREF(bs->full_open_options);
3288      bs->full_open_options = NULL;
3289  
3290      bdrv_release_named_dirty_bitmaps(bs);
3291      assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3292  
3293      QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3294          g_free(ban);
3295      }
3296      QLIST_INIT(&bs->aio_notifiers);
3297      bdrv_drained_end(bs);
3298  }
3299  
3300  void bdrv_close_all(void)
3301  {
3302      block_job_cancel_sync_all();
3303      nbd_export_close_all();
3304  
3305      /* Drop references from requests still in flight, such as canceled block
3306       * jobs whose AIO context has not been polled yet */
3307      bdrv_drain_all();
3308  
3309      blk_remove_all_bs();
3310      blockdev_close_all_bdrv_states();
3311  
3312      assert(QTAILQ_EMPTY(&all_bdrv_states));
3313  }
3314  
3315  static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3316  {
3317      BdrvChild *to_c;
3318  
3319      if (c->role->stay_at_node) {
3320          return false;
3321      }
3322  
3323      if (c->role == &child_backing) {
3324          /* If @from is a backing file of @to, ignore the child to avoid
3325           * creating a loop. We only want to change the pointer of other
3326           * parents. */
3327          QLIST_FOREACH(to_c, &to->children, next) {
3328              if (to_c == c) {
3329                  break;
3330              }
3331          }
3332          if (to_c) {
3333              return false;
3334          }
3335      }
3336  
3337      return true;
3338  }
3339  
3340  void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3341                         Error **errp)
3342  {
3343      BdrvChild *c, *next;
3344      GSList *list = NULL, *p;
3345      uint64_t old_perm, old_shared;
3346      uint64_t perm = 0, shared = BLK_PERM_ALL;
3347      int ret;
3348  
3349      assert(!atomic_read(&from->in_flight));
3350      assert(!atomic_read(&to->in_flight));
3351  
3352      /* Make sure that @from doesn't go away until we have successfully attached
3353       * all of its parents to @to. */
3354      bdrv_ref(from);
3355  
3356      /* Put all parents into @list and calculate their cumulative permissions */
3357      QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3358          if (!should_update_child(c, to)) {
3359              continue;
3360          }
3361          list = g_slist_prepend(list, c);
3362          perm |= c->perm;
3363          shared &= c->shared_perm;
3364      }
3365  
3366      /* Check whether the required permissions can be granted on @to, ignoring
3367       * all BdrvChild in @list so that they can't block themselves. */
3368      ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
3369      if (ret < 0) {
3370          bdrv_abort_perm_update(to);
3371          goto out;
3372      }
3373  
3374      /* Now actually perform the change. We performed the permission check for
3375       * all elements of @list at once, so set the permissions all at once at the
3376       * very end. */
3377      for (p = list; p != NULL; p = p->next) {
3378          c = p->data;
3379  
3380          bdrv_ref(to);
3381          bdrv_replace_child_noperm(c, to);
3382          bdrv_unref(from);
3383      }
3384  
3385      bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3386      bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3387  
3388  out:
3389      g_slist_free(list);
3390      bdrv_unref(from);
3391  }
3392  
3393  /*
3394   * Add new bs contents at the top of an image chain while the chain is
3395   * live, while keeping required fields on the top layer.
3396   *
3397   * This will modify the BlockDriverState fields, and swap contents
3398   * between bs_new and bs_top. Both bs_new and bs_top are modified.
3399   *
3400   * bs_new must not be attached to a BlockBackend.
3401   *
3402   * This function does not create any image files.
3403   *
3404   * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3405   * that's what the callers commonly need. bs_new will be referenced by the old
3406   * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3407   * reference of its own, it must call bdrv_ref().
3408   */
3409  void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3410                   Error **errp)
3411  {
3412      Error *local_err = NULL;
3413  
3414      bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3415      if (local_err) {
3416          error_propagate(errp, local_err);
3417          goto out;
3418      }
3419  
3420      bdrv_replace_node(bs_top, bs_new, &local_err);
3421      if (local_err) {
3422          error_propagate(errp, local_err);
3423          bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3424          goto out;
3425      }
3426  
3427      /* bs_new is now referenced by its new parents, we don't need the
3428       * additional reference any more. */
3429  out:
3430      bdrv_unref(bs_new);
3431  }
3432  
3433  static void bdrv_delete(BlockDriverState *bs)
3434  {
3435      assert(!bs->job);
3436      assert(bdrv_op_blocker_is_empty(bs));
3437      assert(!bs->refcnt);
3438  
3439      bdrv_close(bs);
3440  
3441      /* remove from list, if necessary */
3442      if (bs->node_name[0] != '\0') {
3443          QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3444      }
3445      QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3446  
3447      g_free(bs);
3448  }
3449  
3450  /*
3451   * Run consistency checks on an image
3452   *
3453   * Returns 0 if the check could be completed (it doesn't mean that the image is
3454   * free of errors) or -errno when an internal error occurred. The results of the
3455   * check are stored in res.
3456   */
3457  int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
3458  {
3459      if (bs->drv == NULL) {
3460          return -ENOMEDIUM;
3461      }
3462      if (bs->drv->bdrv_check == NULL) {
3463          return -ENOTSUP;
3464      }
3465  
3466      memset(res, 0, sizeof(*res));
3467      return bs->drv->bdrv_check(bs, res, fix);
3468  }
3469  
3470  /*
3471   * Return values:
3472   * 0        - success
3473   * -EINVAL  - backing format specified, but no file
3474   * -ENOSPC  - can't update the backing file because no space is left in the
3475   *            image file header
3476   * -ENOTSUP - format driver doesn't support changing the backing file
3477   */
3478  int bdrv_change_backing_file(BlockDriverState *bs,
3479      const char *backing_file, const char *backing_fmt)
3480  {
3481      BlockDriver *drv = bs->drv;
3482      int ret;
3483  
3484      if (!drv) {
3485          return -ENOMEDIUM;
3486      }
3487  
3488      /* Backing file format doesn't make sense without a backing file */
3489      if (backing_fmt && !backing_file) {
3490          return -EINVAL;
3491      }
3492  
3493      if (drv->bdrv_change_backing_file != NULL) {
3494          ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3495      } else {
3496          ret = -ENOTSUP;
3497      }
3498  
3499      if (ret == 0) {
3500          pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3501          pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3502      }
3503      return ret;
3504  }
3505  
3506  /*
3507   * Finds the image layer in the chain that has 'bs' as its backing file.
3508   *
3509   * active is the current topmost image.
3510   *
3511   * Returns NULL if bs is not found in active's image chain,
3512   * or if active == bs.
3513   *
3514   * Returns the bottommost base image if bs == NULL.
3515   */
3516  BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3517                                      BlockDriverState *bs)
3518  {
3519      while (active && bs != backing_bs(active)) {
3520          active = backing_bs(active);
3521      }
3522  
3523      return active;
3524  }
3525  
3526  /* Given a BDS, searches for the base layer. */
3527  BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3528  {
3529      return bdrv_find_overlay(bs, NULL);
3530  }
3531  
3532  /*
3533   * Drops images above 'base' up to and including 'top', and sets the image
3534   * above 'top' to have base as its backing file.
3535   *
3536   * Requires that the overlay to 'top' is opened r/w, so that the backing file
3537   * information in 'bs' can be properly updated.
3538   *
3539   * E.g., this will convert the following chain:
3540   * bottom <- base <- intermediate <- top <- active
3541   *
3542   * to
3543   *
3544   * bottom <- base <- active
3545   *
3546   * It is allowed for bottom==base, in which case it converts:
3547   *
3548   * base <- intermediate <- top <- active
3549   *
3550   * to
3551   *
3552   * base <- active
3553   *
3554   * If backing_file_str is non-NULL, it will be used when modifying top's
3555   * overlay image metadata.
3556   *
3557   * Error conditions:
3558   *  if active == top, that is considered an error
3559   *
3560   */
3561  int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
3562                             const char *backing_file_str)
3563  {
3564      BdrvChild *c, *next;
3565      Error *local_err = NULL;
3566      int ret = -EIO;
3567  
3568      bdrv_ref(top);
3569  
3570      if (!top->drv || !base->drv) {
3571          goto exit;
3572      }
3573  
3574      /* Make sure that base is in the backing chain of top */
3575      if (!bdrv_chain_contains(top, base)) {
3576          goto exit;
3577      }
3578  
3579      /* success - we can delete the intermediate states, and link top->base */
3580      /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3581       * we've figured out how they should work. */
3582      backing_file_str = backing_file_str ? backing_file_str : base->filename;
3583  
3584      QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
3585          /* Check whether we are allowed to switch c from top to base */
3586          GSList *ignore_children = g_slist_prepend(NULL, c);
3587          bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
3588                                 ignore_children, &local_err);
3589          if (local_err) {
3590              ret = -EPERM;
3591              error_report_err(local_err);
3592              goto exit;
3593          }
3594          g_slist_free(ignore_children);
3595  
3596          /* If so, update the backing file path in the image file */
3597          if (c->role->update_filename) {
3598              ret = c->role->update_filename(c, base, backing_file_str,
3599                                             &local_err);
3600              if (ret < 0) {
3601                  bdrv_abort_perm_update(base);
3602                  error_report_err(local_err);
3603                  goto exit;
3604              }
3605          }
3606  
3607          /* Do the actual switch in the in-memory graph.
3608           * Completes bdrv_check_update_perm() transaction internally. */
3609          bdrv_ref(base);
3610          bdrv_replace_child(c, base);
3611          bdrv_unref(top);
3612      }
3613  
3614      ret = 0;
3615  exit:
3616      bdrv_unref(top);
3617      return ret;
3618  }
3619  
3620  /**
3621   * Truncate file to 'offset' bytes (needed only for file protocols)
3622   */
3623  int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc,
3624                    Error **errp)
3625  {
3626      BlockDriverState *bs = child->bs;
3627      BlockDriver *drv = bs->drv;
3628      int ret;
3629  
3630      assert(child->perm & BLK_PERM_RESIZE);
3631  
3632      /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3633      if (!drv) {
3634          error_setg(errp, "No medium inserted");
3635          return -ENOMEDIUM;
3636      }
3637      if (!drv->bdrv_truncate) {
3638          if (bs->file && drv->is_filter) {
3639              return bdrv_truncate(bs->file, offset, prealloc, errp);
3640          }
3641          error_setg(errp, "Image format driver does not support resize");
3642          return -ENOTSUP;
3643      }
3644      if (bs->read_only) {
3645          error_setg(errp, "Image is read-only");
3646          return -EACCES;
3647      }
3648  
3649      assert(!(bs->open_flags & BDRV_O_INACTIVE));
3650  
3651      ret = drv->bdrv_truncate(bs, offset, prealloc, errp);
3652      if (ret < 0) {
3653          return ret;
3654      }
3655      ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3656      if (ret < 0) {
3657          error_setg_errno(errp, -ret, "Could not refresh total sector count");
3658      } else {
3659          offset = bs->total_sectors * BDRV_SECTOR_SIZE;
3660      }
3661      bdrv_dirty_bitmap_truncate(bs, offset);
3662      bdrv_parent_cb_resize(bs);
3663      atomic_inc(&bs->write_gen);
3664      return ret;
3665  }
3666  
3667  /**
3668   * Length of a allocated file in bytes. Sparse files are counted by actual
3669   * allocated space. Return < 0 if error or unknown.
3670   */
3671  int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3672  {
3673      BlockDriver *drv = bs->drv;
3674      if (!drv) {
3675          return -ENOMEDIUM;
3676      }
3677      if (drv->bdrv_get_allocated_file_size) {
3678          return drv->bdrv_get_allocated_file_size(bs);
3679      }
3680      if (bs->file) {
3681          return bdrv_get_allocated_file_size(bs->file->bs);
3682      }
3683      return -ENOTSUP;
3684  }
3685  
3686  /*
3687   * bdrv_measure:
3688   * @drv: Format driver
3689   * @opts: Creation options for new image
3690   * @in_bs: Existing image containing data for new image (may be NULL)
3691   * @errp: Error object
3692   * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3693   *          or NULL on error
3694   *
3695   * Calculate file size required to create a new image.
3696   *
3697   * If @in_bs is given then space for allocated clusters and zero clusters
3698   * from that image are included in the calculation.  If @opts contains a
3699   * backing file that is shared by @in_bs then backing clusters may be omitted
3700   * from the calculation.
3701   *
3702   * If @in_bs is NULL then the calculation includes no allocated clusters
3703   * unless a preallocation option is given in @opts.
3704   *
3705   * Note that @in_bs may use a different BlockDriver from @drv.
3706   *
3707   * If an error occurs the @errp pointer is set.
3708   */
3709  BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
3710                                 BlockDriverState *in_bs, Error **errp)
3711  {
3712      if (!drv->bdrv_measure) {
3713          error_setg(errp, "Block driver '%s' does not support size measurement",
3714                     drv->format_name);
3715          return NULL;
3716      }
3717  
3718      return drv->bdrv_measure(opts, in_bs, errp);
3719  }
3720  
3721  /**
3722   * Return number of sectors on success, -errno on error.
3723   */
3724  int64_t bdrv_nb_sectors(BlockDriverState *bs)
3725  {
3726      BlockDriver *drv = bs->drv;
3727  
3728      if (!drv)
3729          return -ENOMEDIUM;
3730  
3731      if (drv->has_variable_length) {
3732          int ret = refresh_total_sectors(bs, bs->total_sectors);
3733          if (ret < 0) {
3734              return ret;
3735          }
3736      }
3737      return bs->total_sectors;
3738  }
3739  
3740  /**
3741   * Return length in bytes on success, -errno on error.
3742   * The length is always a multiple of BDRV_SECTOR_SIZE.
3743   */
3744  int64_t bdrv_getlength(BlockDriverState *bs)
3745  {
3746      int64_t ret = bdrv_nb_sectors(bs);
3747  
3748      ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3749      return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3750  }
3751  
3752  /* return 0 as number of sectors if no device present or error */
3753  void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3754  {
3755      int64_t nb_sectors = bdrv_nb_sectors(bs);
3756  
3757      *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3758  }
3759  
3760  bool bdrv_is_sg(BlockDriverState *bs)
3761  {
3762      return bs->sg;
3763  }
3764  
3765  bool bdrv_is_encrypted(BlockDriverState *bs)
3766  {
3767      if (bs->backing && bs->backing->bs->encrypted) {
3768          return true;
3769      }
3770      return bs->encrypted;
3771  }
3772  
3773  const char *bdrv_get_format_name(BlockDriverState *bs)
3774  {
3775      return bs->drv ? bs->drv->format_name : NULL;
3776  }
3777  
3778  static int qsort_strcmp(const void *a, const void *b)
3779  {
3780      return strcmp(*(char *const *)a, *(char *const *)b);
3781  }
3782  
3783  void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3784                           void *opaque)
3785  {
3786      BlockDriver *drv;
3787      int count = 0;
3788      int i;
3789      const char **formats = NULL;
3790  
3791      QLIST_FOREACH(drv, &bdrv_drivers, list) {
3792          if (drv->format_name) {
3793              bool found = false;
3794              int i = count;
3795              while (formats && i && !found) {
3796                  found = !strcmp(formats[--i], drv->format_name);
3797              }
3798  
3799              if (!found) {
3800                  formats = g_renew(const char *, formats, count + 1);
3801                  formats[count++] = drv->format_name;
3802              }
3803          }
3804      }
3805  
3806      for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3807          const char *format_name = block_driver_modules[i].format_name;
3808  
3809          if (format_name) {
3810              bool found = false;
3811              int j = count;
3812  
3813              while (formats && j && !found) {
3814                  found = !strcmp(formats[--j], format_name);
3815              }
3816  
3817              if (!found) {
3818                  formats = g_renew(const char *, formats, count + 1);
3819                  formats[count++] = format_name;
3820              }
3821          }
3822      }
3823  
3824      qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3825  
3826      for (i = 0; i < count; i++) {
3827          it(opaque, formats[i]);
3828      }
3829  
3830      g_free(formats);
3831  }
3832  
3833  /* This function is to find a node in the bs graph */
3834  BlockDriverState *bdrv_find_node(const char *node_name)
3835  {
3836      BlockDriverState *bs;
3837  
3838      assert(node_name);
3839  
3840      QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3841          if (!strcmp(node_name, bs->node_name)) {
3842              return bs;
3843          }
3844      }
3845      return NULL;
3846  }
3847  
3848  /* Put this QMP function here so it can access the static graph_bdrv_states. */
3849  BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3850  {
3851      BlockDeviceInfoList *list, *entry;
3852      BlockDriverState *bs;
3853  
3854      list = NULL;
3855      QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3856          BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3857          if (!info) {
3858              qapi_free_BlockDeviceInfoList(list);
3859              return NULL;
3860          }
3861          entry = g_malloc0(sizeof(*entry));
3862          entry->value = info;
3863          entry->next = list;
3864          list = entry;
3865      }
3866  
3867      return list;
3868  }
3869  
3870  BlockDriverState *bdrv_lookup_bs(const char *device,
3871                                   const char *node_name,
3872                                   Error **errp)
3873  {
3874      BlockBackend *blk;
3875      BlockDriverState *bs;
3876  
3877      if (device) {
3878          blk = blk_by_name(device);
3879  
3880          if (blk) {
3881              bs = blk_bs(blk);
3882              if (!bs) {
3883                  error_setg(errp, "Device '%s' has no medium", device);
3884              }
3885  
3886              return bs;
3887          }
3888      }
3889  
3890      if (node_name) {
3891          bs = bdrv_find_node(node_name);
3892  
3893          if (bs) {
3894              return bs;
3895          }
3896      }
3897  
3898      error_setg(errp, "Cannot find device=%s nor node_name=%s",
3899                       device ? device : "",
3900                       node_name ? node_name : "");
3901      return NULL;
3902  }
3903  
3904  /* If 'base' is in the same chain as 'top', return true. Otherwise,
3905   * return false.  If either argument is NULL, return false. */
3906  bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3907  {
3908      while (top && top != base) {
3909          top = backing_bs(top);
3910      }
3911  
3912      return top != NULL;
3913  }
3914  
3915  BlockDriverState *bdrv_next_node(BlockDriverState *bs)
3916  {
3917      if (!bs) {
3918          return QTAILQ_FIRST(&graph_bdrv_states);
3919      }
3920      return QTAILQ_NEXT(bs, node_list);
3921  }
3922  
3923  const char *bdrv_get_node_name(const BlockDriverState *bs)
3924  {
3925      return bs->node_name;
3926  }
3927  
3928  const char *bdrv_get_parent_name(const BlockDriverState *bs)
3929  {
3930      BdrvChild *c;
3931      const char *name;
3932  
3933      /* If multiple parents have a name, just pick the first one. */
3934      QLIST_FOREACH(c, &bs->parents, next_parent) {
3935          if (c->role->get_name) {
3936              name = c->role->get_name(c);
3937              if (name && *name) {
3938                  return name;
3939              }
3940          }
3941      }
3942  
3943      return NULL;
3944  }
3945  
3946  /* TODO check what callers really want: bs->node_name or blk_name() */
3947  const char *bdrv_get_device_name(const BlockDriverState *bs)
3948  {
3949      return bdrv_get_parent_name(bs) ?: "";
3950  }
3951  
3952  /* This can be used to identify nodes that might not have a device
3953   * name associated. Since node and device names live in the same
3954   * namespace, the result is unambiguous. The exception is if both are
3955   * absent, then this returns an empty (non-null) string. */
3956  const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
3957  {
3958      return bdrv_get_parent_name(bs) ?: bs->node_name;
3959  }
3960  
3961  int bdrv_get_flags(BlockDriverState *bs)
3962  {
3963      return bs->open_flags;
3964  }
3965  
3966  int bdrv_has_zero_init_1(BlockDriverState *bs)
3967  {
3968      return 1;
3969  }
3970  
3971  int bdrv_has_zero_init(BlockDriverState *bs)
3972  {
3973      if (!bs->drv) {
3974          return 0;
3975      }
3976  
3977      /* If BS is a copy on write image, it is initialized to
3978         the contents of the base image, which may not be zeroes.  */
3979      if (bs->backing) {
3980          return 0;
3981      }
3982      if (bs->drv->bdrv_has_zero_init) {
3983          return bs->drv->bdrv_has_zero_init(bs);
3984      }
3985      if (bs->file && bs->drv->is_filter) {
3986          return bdrv_has_zero_init(bs->file->bs);
3987      }
3988  
3989      /* safe default */
3990      return 0;
3991  }
3992  
3993  bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3994  {
3995      BlockDriverInfo bdi;
3996  
3997      if (bs->backing) {
3998          return false;
3999      }
4000  
4001      if (bdrv_get_info(bs, &bdi) == 0) {
4002          return bdi.unallocated_blocks_are_zero;
4003      }
4004  
4005      return false;
4006  }
4007  
4008  bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
4009  {
4010      BlockDriverInfo bdi;
4011  
4012      if (!(bs->open_flags & BDRV_O_UNMAP)) {
4013          return false;
4014      }
4015  
4016      if (bdrv_get_info(bs, &bdi) == 0) {
4017          return bdi.can_write_zeroes_with_unmap;
4018      }
4019  
4020      return false;
4021  }
4022  
4023  const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
4024  {
4025      if (bs->backing && bs->backing->bs->encrypted)
4026          return bs->backing_file;
4027      else if (bs->encrypted)
4028          return bs->filename;
4029      else
4030          return NULL;
4031  }
4032  
4033  void bdrv_get_backing_filename(BlockDriverState *bs,
4034                                 char *filename, int filename_size)
4035  {
4036      pstrcpy(filename, filename_size, bs->backing_file);
4037  }
4038  
4039  int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4040  {
4041      BlockDriver *drv = bs->drv;
4042      /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4043      if (!drv) {
4044          return -ENOMEDIUM;
4045      }
4046      if (!drv->bdrv_get_info) {
4047          if (bs->file && drv->is_filter) {
4048              return bdrv_get_info(bs->file->bs, bdi);
4049          }
4050          return -ENOTSUP;
4051      }
4052      memset(bdi, 0, sizeof(*bdi));
4053      return drv->bdrv_get_info(bs, bdi);
4054  }
4055  
4056  ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4057  {
4058      BlockDriver *drv = bs->drv;
4059      if (drv && drv->bdrv_get_specific_info) {
4060          return drv->bdrv_get_specific_info(bs);
4061      }
4062      return NULL;
4063  }
4064  
4065  void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4066  {
4067      if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4068          return;
4069      }
4070  
4071      bs->drv->bdrv_debug_event(bs, event);
4072  }
4073  
4074  int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4075                            const char *tag)
4076  {
4077      while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4078          bs = bs->file ? bs->file->bs : NULL;
4079      }
4080  
4081      if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4082          return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4083      }
4084  
4085      return -ENOTSUP;
4086  }
4087  
4088  int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4089  {
4090      while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4091          bs = bs->file ? bs->file->bs : NULL;
4092      }
4093  
4094      if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4095          return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4096      }
4097  
4098      return -ENOTSUP;
4099  }
4100  
4101  int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4102  {
4103      while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4104          bs = bs->file ? bs->file->bs : NULL;
4105      }
4106  
4107      if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4108          return bs->drv->bdrv_debug_resume(bs, tag);
4109      }
4110  
4111      return -ENOTSUP;
4112  }
4113  
4114  bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4115  {
4116      while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4117          bs = bs->file ? bs->file->bs : NULL;
4118      }
4119  
4120      if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4121          return bs->drv->bdrv_debug_is_suspended(bs, tag);
4122      }
4123  
4124      return false;
4125  }
4126  
4127  /* backing_file can either be relative, or absolute, or a protocol.  If it is
4128   * relative, it must be relative to the chain.  So, passing in bs->filename
4129   * from a BDS as backing_file should not be done, as that may be relative to
4130   * the CWD rather than the chain. */
4131  BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4132          const char *backing_file)
4133  {
4134      char *filename_full = NULL;
4135      char *backing_file_full = NULL;
4136      char *filename_tmp = NULL;
4137      int is_protocol = 0;
4138      BlockDriverState *curr_bs = NULL;
4139      BlockDriverState *retval = NULL;
4140      Error *local_error = NULL;
4141  
4142      if (!bs || !bs->drv || !backing_file) {
4143          return NULL;
4144      }
4145  
4146      filename_full     = g_malloc(PATH_MAX);
4147      backing_file_full = g_malloc(PATH_MAX);
4148      filename_tmp      = g_malloc(PATH_MAX);
4149  
4150      is_protocol = path_has_protocol(backing_file);
4151  
4152      for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
4153  
4154          /* If either of the filename paths is actually a protocol, then
4155           * compare unmodified paths; otherwise make paths relative */
4156          if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4157              if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4158                  retval = curr_bs->backing->bs;
4159                  break;
4160              }
4161              /* Also check against the full backing filename for the image */
4162              bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
4163                                             &local_error);
4164              if (local_error == NULL) {
4165                  if (strcmp(backing_file, backing_file_full) == 0) {
4166                      retval = curr_bs->backing->bs;
4167                      break;
4168                  }
4169              } else {
4170                  error_free(local_error);
4171                  local_error = NULL;
4172              }
4173          } else {
4174              /* If not an absolute filename path, make it relative to the current
4175               * image's filename path */
4176              path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4177                           backing_file);
4178  
4179              /* We are going to compare absolute pathnames */
4180              if (!realpath(filename_tmp, filename_full)) {
4181                  continue;
4182              }
4183  
4184              /* We need to make sure the backing filename we are comparing against
4185               * is relative to the current image filename (or absolute) */
4186              path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4187                           curr_bs->backing_file);
4188  
4189              if (!realpath(filename_tmp, backing_file_full)) {
4190                  continue;
4191              }
4192  
4193              if (strcmp(backing_file_full, filename_full) == 0) {
4194                  retval = curr_bs->backing->bs;
4195                  break;
4196              }
4197          }
4198      }
4199  
4200      g_free(filename_full);
4201      g_free(backing_file_full);
4202      g_free(filename_tmp);
4203      return retval;
4204  }
4205  
4206  void bdrv_init(void)
4207  {
4208      module_call_init(MODULE_INIT_BLOCK);
4209  }
4210  
4211  void bdrv_init_with_whitelist(void)
4212  {
4213      use_bdrv_whitelist = 1;
4214      bdrv_init();
4215  }
4216  
4217  void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4218  {
4219      BdrvChild *child, *parent;
4220      uint64_t perm, shared_perm;
4221      Error *local_err = NULL;
4222      int ret;
4223  
4224      if (!bs->drv)  {
4225          return;
4226      }
4227  
4228      if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4229          return;
4230      }
4231  
4232      QLIST_FOREACH(child, &bs->children, next) {
4233          bdrv_invalidate_cache(child->bs, &local_err);
4234          if (local_err) {
4235              error_propagate(errp, local_err);
4236              return;
4237          }
4238      }
4239  
4240      /*
4241       * Update permissions, they may differ for inactive nodes.
4242       *
4243       * Note that the required permissions of inactive images are always a
4244       * subset of the permissions required after activating the image. This
4245       * allows us to just get the permissions upfront without restricting
4246       * drv->bdrv_invalidate_cache().
4247       *
4248       * It also means that in error cases, we don't have to try and revert to
4249       * the old permissions (which is an operation that could fail, too). We can
4250       * just keep the extended permissions for the next time that an activation
4251       * of the image is tried.
4252       */
4253      bs->open_flags &= ~BDRV_O_INACTIVE;
4254      bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4255      ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4256      if (ret < 0) {
4257          bs->open_flags |= BDRV_O_INACTIVE;
4258          error_propagate(errp, local_err);
4259          return;
4260      }
4261      bdrv_set_perm(bs, perm, shared_perm);
4262  
4263      if (bs->drv->bdrv_invalidate_cache) {
4264          bs->drv->bdrv_invalidate_cache(bs, &local_err);
4265          if (local_err) {
4266              bs->open_flags |= BDRV_O_INACTIVE;
4267              error_propagate(errp, local_err);
4268              return;
4269          }
4270      }
4271  
4272      ret = refresh_total_sectors(bs, bs->total_sectors);
4273      if (ret < 0) {
4274          bs->open_flags |= BDRV_O_INACTIVE;
4275          error_setg_errno(errp, -ret, "Could not refresh total sector count");
4276          return;
4277      }
4278  
4279      QLIST_FOREACH(parent, &bs->parents, next_parent) {
4280          if (parent->role->activate) {
4281              parent->role->activate(parent, &local_err);
4282              if (local_err) {
4283                  error_propagate(errp, local_err);
4284                  return;
4285              }
4286          }
4287      }
4288  }
4289  
4290  void bdrv_invalidate_cache_all(Error **errp)
4291  {
4292      BlockDriverState *bs;
4293      Error *local_err = NULL;
4294      BdrvNextIterator it;
4295  
4296      for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4297          AioContext *aio_context = bdrv_get_aio_context(bs);
4298  
4299          aio_context_acquire(aio_context);
4300          bdrv_invalidate_cache(bs, &local_err);
4301          aio_context_release(aio_context);
4302          if (local_err) {
4303              error_propagate(errp, local_err);
4304              bdrv_next_cleanup(&it);
4305              return;
4306          }
4307      }
4308  }
4309  
4310  static int bdrv_inactivate_recurse(BlockDriverState *bs,
4311                                     bool setting_flag)
4312  {
4313      BdrvChild *child, *parent;
4314      int ret;
4315  
4316      if (!bs->drv) {
4317          return -ENOMEDIUM;
4318      }
4319  
4320      if (!setting_flag && bs->drv->bdrv_inactivate) {
4321          ret = bs->drv->bdrv_inactivate(bs);
4322          if (ret < 0) {
4323              return ret;
4324          }
4325      }
4326  
4327      if (setting_flag && !(bs->open_flags & BDRV_O_INACTIVE)) {
4328          uint64_t perm, shared_perm;
4329  
4330          QLIST_FOREACH(parent, &bs->parents, next_parent) {
4331              if (parent->role->inactivate) {
4332                  ret = parent->role->inactivate(parent);
4333                  if (ret < 0) {
4334                      return ret;
4335                  }
4336              }
4337          }
4338  
4339          bs->open_flags |= BDRV_O_INACTIVE;
4340  
4341          /* Update permissions, they may differ for inactive nodes */
4342          bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4343          bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
4344          bdrv_set_perm(bs, perm, shared_perm);
4345      }
4346  
4347      QLIST_FOREACH(child, &bs->children, next) {
4348          ret = bdrv_inactivate_recurse(child->bs, setting_flag);
4349          if (ret < 0) {
4350              return ret;
4351          }
4352      }
4353  
4354      /* At this point persistent bitmaps should be already stored by the format
4355       * driver */
4356      bdrv_release_persistent_dirty_bitmaps(bs);
4357  
4358      return 0;
4359  }
4360  
4361  int bdrv_inactivate_all(void)
4362  {
4363      BlockDriverState *bs = NULL;
4364      BdrvNextIterator it;
4365      int ret = 0;
4366      int pass;
4367      GSList *aio_ctxs = NULL, *ctx;
4368  
4369      for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4370          AioContext *aio_context = bdrv_get_aio_context(bs);
4371  
4372          if (!g_slist_find(aio_ctxs, aio_context)) {
4373              aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
4374              aio_context_acquire(aio_context);
4375          }
4376      }
4377  
4378      /* We do two passes of inactivation. The first pass calls to drivers'
4379       * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4380       * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4381       * is allowed. */
4382      for (pass = 0; pass < 2; pass++) {
4383          for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4384              ret = bdrv_inactivate_recurse(bs, pass);
4385              if (ret < 0) {
4386                  bdrv_next_cleanup(&it);
4387                  goto out;
4388              }
4389          }
4390      }
4391  
4392  out:
4393      for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
4394          AioContext *aio_context = ctx->data;
4395          aio_context_release(aio_context);
4396      }
4397      g_slist_free(aio_ctxs);
4398  
4399      return ret;
4400  }
4401  
4402  /**************************************************************/
4403  /* removable device support */
4404  
4405  /**
4406   * Return TRUE if the media is present
4407   */
4408  bool bdrv_is_inserted(BlockDriverState *bs)
4409  {
4410      BlockDriver *drv = bs->drv;
4411      BdrvChild *child;
4412  
4413      if (!drv) {
4414          return false;
4415      }
4416      if (drv->bdrv_is_inserted) {
4417          return drv->bdrv_is_inserted(bs);
4418      }
4419      QLIST_FOREACH(child, &bs->children, next) {
4420          if (!bdrv_is_inserted(child->bs)) {
4421              return false;
4422          }
4423      }
4424      return true;
4425  }
4426  
4427  /**
4428   * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4429   */
4430  void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4431  {
4432      BlockDriver *drv = bs->drv;
4433  
4434      if (drv && drv->bdrv_eject) {
4435          drv->bdrv_eject(bs, eject_flag);
4436      }
4437  }
4438  
4439  /**
4440   * Lock or unlock the media (if it is locked, the user won't be able
4441   * to eject it manually).
4442   */
4443  void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4444  {
4445      BlockDriver *drv = bs->drv;
4446  
4447      trace_bdrv_lock_medium(bs, locked);
4448  
4449      if (drv && drv->bdrv_lock_medium) {
4450          drv->bdrv_lock_medium(bs, locked);
4451      }
4452  }
4453  
4454  /* Get a reference to bs */
4455  void bdrv_ref(BlockDriverState *bs)
4456  {
4457      bs->refcnt++;
4458  }
4459  
4460  /* Release a previously grabbed reference to bs.
4461   * If after releasing, reference count is zero, the BlockDriverState is
4462   * deleted. */
4463  void bdrv_unref(BlockDriverState *bs)
4464  {
4465      if (!bs) {
4466          return;
4467      }
4468      assert(bs->refcnt > 0);
4469      if (--bs->refcnt == 0) {
4470          bdrv_delete(bs);
4471      }
4472  }
4473  
4474  struct BdrvOpBlocker {
4475      Error *reason;
4476      QLIST_ENTRY(BdrvOpBlocker) list;
4477  };
4478  
4479  bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4480  {
4481      BdrvOpBlocker *blocker;
4482      assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4483      if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4484          blocker = QLIST_FIRST(&bs->op_blockers[op]);
4485          error_propagate(errp, error_copy(blocker->reason));
4486          error_prepend(errp, "Node '%s' is busy: ",
4487                        bdrv_get_device_or_node_name(bs));
4488          return true;
4489      }
4490      return false;
4491  }
4492  
4493  void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4494  {
4495      BdrvOpBlocker *blocker;
4496      assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4497  
4498      blocker = g_new0(BdrvOpBlocker, 1);
4499      blocker->reason = reason;
4500      QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4501  }
4502  
4503  void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4504  {
4505      BdrvOpBlocker *blocker, *next;
4506      assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4507      QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4508          if (blocker->reason == reason) {
4509              QLIST_REMOVE(blocker, list);
4510              g_free(blocker);
4511          }
4512      }
4513  }
4514  
4515  void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4516  {
4517      int i;
4518      for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4519          bdrv_op_block(bs, i, reason);
4520      }
4521  }
4522  
4523  void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4524  {
4525      int i;
4526      for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4527          bdrv_op_unblock(bs, i, reason);
4528      }
4529  }
4530  
4531  bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4532  {
4533      int i;
4534  
4535      for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4536          if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4537              return false;
4538          }
4539      }
4540      return true;
4541  }
4542  
4543  void bdrv_img_create(const char *filename, const char *fmt,
4544                       const char *base_filename, const char *base_fmt,
4545                       char *options, uint64_t img_size, int flags, bool quiet,
4546                       Error **errp)
4547  {
4548      QemuOptsList *create_opts = NULL;
4549      QemuOpts *opts = NULL;
4550      const char *backing_fmt, *backing_file;
4551      int64_t size;
4552      BlockDriver *drv, *proto_drv;
4553      Error *local_err = NULL;
4554      int ret = 0;
4555  
4556      /* Find driver and parse its options */
4557      drv = bdrv_find_format(fmt);
4558      if (!drv) {
4559          error_setg(errp, "Unknown file format '%s'", fmt);
4560          return;
4561      }
4562  
4563      proto_drv = bdrv_find_protocol(filename, true, errp);
4564      if (!proto_drv) {
4565          return;
4566      }
4567  
4568      if (!drv->create_opts) {
4569          error_setg(errp, "Format driver '%s' does not support image creation",
4570                     drv->format_name);
4571          return;
4572      }
4573  
4574      if (!proto_drv->create_opts) {
4575          error_setg(errp, "Protocol driver '%s' does not support image creation",
4576                     proto_drv->format_name);
4577          return;
4578      }
4579  
4580      create_opts = qemu_opts_append(create_opts, drv->create_opts);
4581      create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4582  
4583      /* Create parameter list with default values */
4584      opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4585      qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4586  
4587      /* Parse -o options */
4588      if (options) {
4589          qemu_opts_do_parse(opts, options, NULL, &local_err);
4590          if (local_err) {
4591              error_report_err(local_err);
4592              local_err = NULL;
4593              error_setg(errp, "Invalid options for file format '%s'", fmt);
4594              goto out;
4595          }
4596      }
4597  
4598      if (base_filename) {
4599          qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4600          if (local_err) {
4601              error_setg(errp, "Backing file not supported for file format '%s'",
4602                         fmt);
4603              goto out;
4604          }
4605      }
4606  
4607      if (base_fmt) {
4608          qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4609          if (local_err) {
4610              error_setg(errp, "Backing file format not supported for file "
4611                               "format '%s'", fmt);
4612              goto out;
4613          }
4614      }
4615  
4616      backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4617      if (backing_file) {
4618          if (!strcmp(filename, backing_file)) {
4619              error_setg(errp, "Error: Trying to create an image with the "
4620                               "same filename as the backing file");
4621              goto out;
4622          }
4623      }
4624  
4625      backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4626  
4627      /* The size for the image must always be specified, unless we have a backing
4628       * file and we have not been forbidden from opening it. */
4629      size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
4630      if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
4631          BlockDriverState *bs;
4632          char *full_backing = g_new0(char, PATH_MAX);
4633          int back_flags;
4634          QDict *backing_options = NULL;
4635  
4636          bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4637                                                       full_backing, PATH_MAX,
4638                                                       &local_err);
4639          if (local_err) {
4640              g_free(full_backing);
4641              goto out;
4642          }
4643  
4644          /* backing files always opened read-only */
4645          back_flags = flags;
4646          back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4647  
4648          backing_options = qdict_new();
4649          if (backing_fmt) {
4650              qdict_put_str(backing_options, "driver", backing_fmt);
4651          }
4652          qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
4653  
4654          bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4655                         &local_err);
4656          g_free(full_backing);
4657          if (!bs && size != -1) {
4658              /* Couldn't open BS, but we have a size, so it's nonfatal */
4659              warn_reportf_err(local_err,
4660                              "Could not verify backing image. "
4661                              "This may become an error in future versions.\n");
4662              local_err = NULL;
4663          } else if (!bs) {
4664              /* Couldn't open bs, do not have size */
4665              error_append_hint(&local_err,
4666                                "Could not open backing image to determine size.\n");
4667              goto out;
4668          } else {
4669              if (size == -1) {
4670                  /* Opened BS, have no size */
4671                  size = bdrv_getlength(bs);
4672                  if (size < 0) {
4673                      error_setg_errno(errp, -size, "Could not get size of '%s'",
4674                                       backing_file);
4675                      bdrv_unref(bs);
4676                      goto out;
4677                  }
4678                  qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4679              }
4680              bdrv_unref(bs);
4681          }
4682      } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4683  
4684      if (size == -1) {
4685          error_setg(errp, "Image creation needs a size parameter");
4686          goto out;
4687      }
4688  
4689      if (!quiet) {
4690          printf("Formatting '%s', fmt=%s ", filename, fmt);
4691          qemu_opts_print(opts, " ");
4692          puts("");
4693      }
4694  
4695      ret = bdrv_create(drv, filename, opts, &local_err);
4696  
4697      if (ret == -EFBIG) {
4698          /* This is generally a better message than whatever the driver would
4699           * deliver (especially because of the cluster_size_hint), since that
4700           * is most probably not much different from "image too large". */
4701          const char *cluster_size_hint = "";
4702          if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4703              cluster_size_hint = " (try using a larger cluster size)";
4704          }
4705          error_setg(errp, "The image size is too large for file format '%s'"
4706                     "%s", fmt, cluster_size_hint);
4707          error_free(local_err);
4708          local_err = NULL;
4709      }
4710  
4711  out:
4712      qemu_opts_del(opts);
4713      qemu_opts_free(create_opts);
4714      error_propagate(errp, local_err);
4715  }
4716  
4717  AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4718  {
4719      return bs->aio_context;
4720  }
4721  
4722  void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4723  {
4724      aio_co_enter(bdrv_get_aio_context(bs), co);
4725  }
4726  
4727  static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4728  {
4729      QLIST_REMOVE(ban, list);
4730      g_free(ban);
4731  }
4732  
4733  void bdrv_detach_aio_context(BlockDriverState *bs)
4734  {
4735      BdrvAioNotifier *baf, *baf_tmp;
4736      BdrvChild *child;
4737  
4738      if (!bs->drv) {
4739          return;
4740      }
4741  
4742      assert(!bs->walking_aio_notifiers);
4743      bs->walking_aio_notifiers = true;
4744      QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4745          if (baf->deleted) {
4746              bdrv_do_remove_aio_context_notifier(baf);
4747          } else {
4748              baf->detach_aio_context(baf->opaque);
4749          }
4750      }
4751      /* Never mind iterating again to check for ->deleted.  bdrv_close() will
4752       * remove remaining aio notifiers if we aren't called again.
4753       */
4754      bs->walking_aio_notifiers = false;
4755  
4756      if (bs->drv->bdrv_detach_aio_context) {
4757          bs->drv->bdrv_detach_aio_context(bs);
4758      }
4759      QLIST_FOREACH(child, &bs->children, next) {
4760          bdrv_detach_aio_context(child->bs);
4761      }
4762  
4763      bs->aio_context = NULL;
4764  }
4765  
4766  void bdrv_attach_aio_context(BlockDriverState *bs,
4767                               AioContext *new_context)
4768  {
4769      BdrvAioNotifier *ban, *ban_tmp;
4770      BdrvChild *child;
4771  
4772      if (!bs->drv) {
4773          return;
4774      }
4775  
4776      bs->aio_context = new_context;
4777  
4778      QLIST_FOREACH(child, &bs->children, next) {
4779          bdrv_attach_aio_context(child->bs, new_context);
4780      }
4781      if (bs->drv->bdrv_attach_aio_context) {
4782          bs->drv->bdrv_attach_aio_context(bs, new_context);
4783      }
4784  
4785      assert(!bs->walking_aio_notifiers);
4786      bs->walking_aio_notifiers = true;
4787      QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4788          if (ban->deleted) {
4789              bdrv_do_remove_aio_context_notifier(ban);
4790          } else {
4791              ban->attached_aio_context(new_context, ban->opaque);
4792          }
4793      }
4794      bs->walking_aio_notifiers = false;
4795  }
4796  
4797  void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4798  {
4799      AioContext *ctx = bdrv_get_aio_context(bs);
4800  
4801      aio_disable_external(ctx);
4802      bdrv_parent_drained_begin(bs, NULL);
4803      bdrv_drain(bs); /* ensure there are no in-flight requests */
4804  
4805      while (aio_poll(ctx, false)) {
4806          /* wait for all bottom halves to execute */
4807      }
4808  
4809      bdrv_detach_aio_context(bs);
4810  
4811      /* This function executes in the old AioContext so acquire the new one in
4812       * case it runs in a different thread.
4813       */
4814      aio_context_acquire(new_context);
4815      bdrv_attach_aio_context(bs, new_context);
4816      bdrv_parent_drained_end(bs, NULL);
4817      aio_enable_external(ctx);
4818      aio_context_release(new_context);
4819  }
4820  
4821  void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4822          void (*attached_aio_context)(AioContext *new_context, void *opaque),
4823          void (*detach_aio_context)(void *opaque), void *opaque)
4824  {
4825      BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4826      *ban = (BdrvAioNotifier){
4827          .attached_aio_context = attached_aio_context,
4828          .detach_aio_context   = detach_aio_context,
4829          .opaque               = opaque
4830      };
4831  
4832      QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
4833  }
4834  
4835  void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
4836                                        void (*attached_aio_context)(AioContext *,
4837                                                                     void *),
4838                                        void (*detach_aio_context)(void *),
4839                                        void *opaque)
4840  {
4841      BdrvAioNotifier *ban, *ban_next;
4842  
4843      QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4844          if (ban->attached_aio_context == attached_aio_context &&
4845              ban->detach_aio_context   == detach_aio_context   &&
4846              ban->opaque               == opaque               &&
4847              ban->deleted              == false)
4848          {
4849              if (bs->walking_aio_notifiers) {
4850                  ban->deleted = true;
4851              } else {
4852                  bdrv_do_remove_aio_context_notifier(ban);
4853              }
4854              return;
4855          }
4856      }
4857  
4858      abort();
4859  }
4860  
4861  int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
4862                         BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
4863  {
4864      if (!bs->drv) {
4865          return -ENOMEDIUM;
4866      }
4867      if (!bs->drv->bdrv_amend_options) {
4868          return -ENOTSUP;
4869      }
4870      return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
4871  }
4872  
4873  /* This function will be called by the bdrv_recurse_is_first_non_filter method
4874   * of block filter and by bdrv_is_first_non_filter.
4875   * It is used to test if the given bs is the candidate or recurse more in the
4876   * node graph.
4877   */
4878  bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
4879                                        BlockDriverState *candidate)
4880  {
4881      /* return false if basic checks fails */
4882      if (!bs || !bs->drv) {
4883          return false;
4884      }
4885  
4886      /* the code reached a non block filter driver -> check if the bs is
4887       * the same as the candidate. It's the recursion termination condition.
4888       */
4889      if (!bs->drv->is_filter) {
4890          return bs == candidate;
4891      }
4892      /* Down this path the driver is a block filter driver */
4893  
4894      /* If the block filter recursion method is defined use it to recurse down
4895       * the node graph.
4896       */
4897      if (bs->drv->bdrv_recurse_is_first_non_filter) {
4898          return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
4899      }
4900  
4901      /* the driver is a block filter but don't allow to recurse -> return false
4902       */
4903      return false;
4904  }
4905  
4906  /* This function checks if the candidate is the first non filter bs down it's
4907   * bs chain. Since we don't have pointers to parents it explore all bs chains
4908   * from the top. Some filters can choose not to pass down the recursion.
4909   */
4910  bool bdrv_is_first_non_filter(BlockDriverState *candidate)
4911  {
4912      BlockDriverState *bs;
4913      BdrvNextIterator it;
4914  
4915      /* walk down the bs forest recursively */
4916      for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4917          bool perm;
4918  
4919          /* try to recurse in this top level bs */
4920          perm = bdrv_recurse_is_first_non_filter(bs, candidate);
4921  
4922          /* candidate is the first non filter */
4923          if (perm) {
4924              bdrv_next_cleanup(&it);
4925              return true;
4926          }
4927      }
4928  
4929      return false;
4930  }
4931  
4932  BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
4933                                          const char *node_name, Error **errp)
4934  {
4935      BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
4936      AioContext *aio_context;
4937  
4938      if (!to_replace_bs) {
4939          error_setg(errp, "Node name '%s' not found", node_name);
4940          return NULL;
4941      }
4942  
4943      aio_context = bdrv_get_aio_context(to_replace_bs);
4944      aio_context_acquire(aio_context);
4945  
4946      if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
4947          to_replace_bs = NULL;
4948          goto out;
4949      }
4950  
4951      /* We don't want arbitrary node of the BDS chain to be replaced only the top
4952       * most non filter in order to prevent data corruption.
4953       * Another benefit is that this tests exclude backing files which are
4954       * blocked by the backing blockers.
4955       */
4956      if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
4957          error_setg(errp, "Only top most non filter can be replaced");
4958          to_replace_bs = NULL;
4959          goto out;
4960      }
4961  
4962  out:
4963      aio_context_release(aio_context);
4964      return to_replace_bs;
4965  }
4966  
4967  static bool append_open_options(QDict *d, BlockDriverState *bs)
4968  {
4969      const QDictEntry *entry;
4970      QemuOptDesc *desc;
4971      BdrvChild *child;
4972      bool found_any = false;
4973      const char *p;
4974  
4975      for (entry = qdict_first(bs->options); entry;
4976           entry = qdict_next(bs->options, entry))
4977      {
4978          /* Exclude options for children */
4979          QLIST_FOREACH(child, &bs->children, next) {
4980              if (strstart(qdict_entry_key(entry), child->name, &p)
4981                  && (!*p || *p == '.'))
4982              {
4983                  break;
4984              }
4985          }
4986          if (child) {
4987              continue;
4988          }
4989  
4990          /* And exclude all non-driver-specific options */
4991          for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
4992              if (!strcmp(qdict_entry_key(entry), desc->name)) {
4993                  break;
4994              }
4995          }
4996          if (desc->name) {
4997              continue;
4998          }
4999  
5000          qobject_incref(qdict_entry_value(entry));
5001          qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
5002          found_any = true;
5003      }
5004  
5005      return found_any;
5006  }
5007  
5008  /* Updates the following BDS fields:
5009   *  - exact_filename: A filename which may be used for opening a block device
5010   *                    which (mostly) equals the given BDS (even without any
5011   *                    other options; so reading and writing must return the same
5012   *                    results, but caching etc. may be different)
5013   *  - full_open_options: Options which, when given when opening a block device
5014   *                       (without a filename), result in a BDS (mostly)
5015   *                       equalling the given one
5016   *  - filename: If exact_filename is set, it is copied here. Otherwise,
5017   *              full_open_options is converted to a JSON object, prefixed with
5018   *              "json:" (for use through the JSON pseudo protocol) and put here.
5019   */
5020  void bdrv_refresh_filename(BlockDriverState *bs)
5021  {
5022      BlockDriver *drv = bs->drv;
5023      QDict *opts;
5024  
5025      if (!drv) {
5026          return;
5027      }
5028  
5029      /* This BDS's file name will most probably depend on its file's name, so
5030       * refresh that first */
5031      if (bs->file) {
5032          bdrv_refresh_filename(bs->file->bs);
5033      }
5034  
5035      if (drv->bdrv_refresh_filename) {
5036          /* Obsolete information is of no use here, so drop the old file name
5037           * information before refreshing it */
5038          bs->exact_filename[0] = '\0';
5039          if (bs->full_open_options) {
5040              QDECREF(bs->full_open_options);
5041              bs->full_open_options = NULL;
5042          }
5043  
5044          opts = qdict_new();
5045          append_open_options(opts, bs);
5046          drv->bdrv_refresh_filename(bs, opts);
5047          QDECREF(opts);
5048      } else if (bs->file) {
5049          /* Try to reconstruct valid information from the underlying file */
5050          bool has_open_options;
5051  
5052          bs->exact_filename[0] = '\0';
5053          if (bs->full_open_options) {
5054              QDECREF(bs->full_open_options);
5055              bs->full_open_options = NULL;
5056          }
5057  
5058          opts = qdict_new();
5059          has_open_options = append_open_options(opts, bs);
5060  
5061          /* If no specific options have been given for this BDS, the filename of
5062           * the underlying file should suffice for this one as well */
5063          if (bs->file->bs->exact_filename[0] && !has_open_options) {
5064              strcpy(bs->exact_filename, bs->file->bs->exact_filename);
5065          }
5066          /* Reconstructing the full options QDict is simple for most format block
5067           * drivers, as long as the full options are known for the underlying
5068           * file BDS. The full options QDict of that file BDS should somehow
5069           * contain a representation of the filename, therefore the following
5070           * suffices without querying the (exact_)filename of this BDS. */
5071          if (bs->file->bs->full_open_options) {
5072              qdict_put_str(opts, "driver", drv->format_name);
5073              QINCREF(bs->file->bs->full_open_options);
5074              qdict_put(opts, "file", bs->file->bs->full_open_options);
5075  
5076              bs->full_open_options = opts;
5077          } else {
5078              QDECREF(opts);
5079          }
5080      } else if (!bs->full_open_options && qdict_size(bs->options)) {
5081          /* There is no underlying file BDS (at least referenced by BDS.file),
5082           * so the full options QDict should be equal to the options given
5083           * specifically for this block device when it was opened (plus the
5084           * driver specification).
5085           * Because those options don't change, there is no need to update
5086           * full_open_options when it's already set. */
5087  
5088          opts = qdict_new();
5089          append_open_options(opts, bs);
5090          qdict_put_str(opts, "driver", drv->format_name);
5091  
5092          if (bs->exact_filename[0]) {
5093              /* This may not work for all block protocol drivers (some may
5094               * require this filename to be parsed), but we have to find some
5095               * default solution here, so just include it. If some block driver
5096               * does not support pure options without any filename at all or
5097               * needs some special format of the options QDict, it needs to
5098               * implement the driver-specific bdrv_refresh_filename() function.
5099               */
5100              qdict_put_str(opts, "filename", bs->exact_filename);
5101          }
5102  
5103          bs->full_open_options = opts;
5104      }
5105  
5106      if (bs->exact_filename[0]) {
5107          pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5108      } else if (bs->full_open_options) {
5109          QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5110          snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5111                   qstring_get_str(json));
5112          QDECREF(json);
5113      }
5114  }
5115  
5116  /*
5117   * Hot add/remove a BDS's child. So the user can take a child offline when
5118   * it is broken and take a new child online
5119   */
5120  void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5121                      Error **errp)
5122  {
5123  
5124      if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5125          error_setg(errp, "The node %s does not support adding a child",
5126                     bdrv_get_device_or_node_name(parent_bs));
5127          return;
5128      }
5129  
5130      if (!QLIST_EMPTY(&child_bs->parents)) {
5131          error_setg(errp, "The node %s already has a parent",
5132                     child_bs->node_name);
5133          return;
5134      }
5135  
5136      parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5137  }
5138  
5139  void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5140  {
5141      BdrvChild *tmp;
5142  
5143      if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5144          error_setg(errp, "The node %s does not support removing a child",
5145                     bdrv_get_device_or_node_name(parent_bs));
5146          return;
5147      }
5148  
5149      QLIST_FOREACH(tmp, &parent_bs->children, next) {
5150          if (tmp == child) {
5151              break;
5152          }
5153      }
5154  
5155      if (!tmp) {
5156          error_setg(errp, "The node %s does not have a child named %s",
5157                     bdrv_get_device_or_node_name(parent_bs),
5158                     bdrv_get_device_or_node_name(child->bs));
5159          return;
5160      }
5161  
5162      parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5163  }
5164  
5165  bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5166                                       uint32_t granularity, Error **errp)
5167  {
5168      BlockDriver *drv = bs->drv;
5169  
5170      if (!drv) {
5171          error_setg_errno(errp, ENOMEDIUM,
5172                           "Can't store persistent bitmaps to %s",
5173                           bdrv_get_device_or_node_name(bs));
5174          return false;
5175      }
5176  
5177      if (!drv->bdrv_can_store_new_dirty_bitmap) {
5178          error_setg_errno(errp, ENOTSUP,
5179                           "Can't store persistent bitmaps to %s",
5180                           bdrv_get_device_or_node_name(bs));
5181          return false;
5182      }
5183  
5184      return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);
5185  }
5186