xref: /openbmc/qemu/qemu-img.c (revision 278f064e)
1 /*
2  * QEMU disk image utility
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 #include <getopt.h>
27 
28 #include "qemu-common.h"
29 #include "qemu-version.h"
30 #include "qapi/error.h"
31 #include "qapi/qapi-commands-block-core.h"
32 #include "qapi/qapi-visit-block-core.h"
33 #include "qapi/qobject-output-visitor.h"
34 #include "qapi/qmp/qjson.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qemu/cutils.h"
37 #include "qemu/config-file.h"
38 #include "qemu/option.h"
39 #include "qemu/error-report.h"
40 #include "qemu/log.h"
41 #include "qemu/main-loop.h"
42 #include "qemu/module.h"
43 #include "qemu/sockets.h"
44 #include "qemu/units.h"
45 #include "qom/object_interfaces.h"
46 #include "sysemu/block-backend.h"
47 #include "block/block_int.h"
48 #include "block/blockjob.h"
49 #include "block/qapi.h"
50 #include "crypto/init.h"
51 #include "trace/control.h"
52 #include "qemu/throttle.h"
53 #include "block/throttle-groups.h"
54 
55 #define QEMU_IMG_VERSION "qemu-img version " QEMU_FULL_VERSION \
56                           "\n" QEMU_COPYRIGHT "\n"
57 
58 typedef struct img_cmd_t {
59     const char *name;
60     int (*handler)(int argc, char **argv);
61 } img_cmd_t;
62 
63 enum {
64     OPTION_OUTPUT = 256,
65     OPTION_BACKING_CHAIN = 257,
66     OPTION_OBJECT = 258,
67     OPTION_IMAGE_OPTS = 259,
68     OPTION_PATTERN = 260,
69     OPTION_FLUSH_INTERVAL = 261,
70     OPTION_NO_DRAIN = 262,
71     OPTION_TARGET_IMAGE_OPTS = 263,
72     OPTION_SIZE = 264,
73     OPTION_PREALLOCATION = 265,
74     OPTION_SHRINK = 266,
75     OPTION_SALVAGE = 267,
76     OPTION_TARGET_IS_ZERO = 268,
77     OPTION_ADD = 269,
78     OPTION_REMOVE = 270,
79     OPTION_CLEAR = 271,
80     OPTION_ENABLE = 272,
81     OPTION_DISABLE = 273,
82     OPTION_MERGE = 274,
83     OPTION_BITMAPS = 275,
84     OPTION_FORCE = 276,
85 };
86 
87 typedef enum OutputFormat {
88     OFORMAT_JSON,
89     OFORMAT_HUMAN,
90 } OutputFormat;
91 
92 /* Default to cache=writeback as data integrity is not important for qemu-img */
93 #define BDRV_DEFAULT_CACHE "writeback"
94 
95 static void format_print(void *opaque, const char *name)
96 {
97     printf(" %s", name);
98 }
99 
100 static void QEMU_NORETURN GCC_FMT_ATTR(1, 2) error_exit(const char *fmt, ...)
101 {
102     va_list ap;
103 
104     va_start(ap, fmt);
105     error_vreport(fmt, ap);
106     va_end(ap);
107 
108     error_printf("Try 'qemu-img --help' for more information\n");
109     exit(EXIT_FAILURE);
110 }
111 
112 static void QEMU_NORETURN missing_argument(const char *option)
113 {
114     error_exit("missing argument for option '%s'", option);
115 }
116 
117 static void QEMU_NORETURN unrecognized_option(const char *option)
118 {
119     error_exit("unrecognized option '%s'", option);
120 }
121 
122 /* Please keep in synch with docs/tools/qemu-img.rst */
123 static void QEMU_NORETURN help(void)
124 {
125     const char *help_msg =
126            QEMU_IMG_VERSION
127            "usage: qemu-img [standard options] command [command options]\n"
128            "QEMU disk image utility\n"
129            "\n"
130            "    '-h', '--help'       display this help and exit\n"
131            "    '-V', '--version'    output version information and exit\n"
132            "    '-T', '--trace'      [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
133            "                         specify tracing options\n"
134            "\n"
135            "Command syntax:\n"
136 #define DEF(option, callback, arg_string)        \
137            "  " arg_string "\n"
138 #include "qemu-img-cmds.h"
139 #undef DEF
140            "\n"
141            "Command parameters:\n"
142            "  'filename' is a disk image filename\n"
143            "  'objectdef' is a QEMU user creatable object definition. See the qemu(1)\n"
144            "    manual page for a description of the object properties. The most common\n"
145            "    object type is a 'secret', which is used to supply passwords and/or\n"
146            "    encryption keys.\n"
147            "  'fmt' is the disk image format. It is guessed automatically in most cases\n"
148            "  'cache' is the cache mode used to write the output disk image, the valid\n"
149            "    options are: 'none', 'writeback' (default, except for convert), 'writethrough',\n"
150            "    'directsync' and 'unsafe' (default for convert)\n"
151            "  'src_cache' is the cache mode used to read input disk images, the valid\n"
152            "    options are the same as for the 'cache' option\n"
153            "  'size' is the disk image size in bytes. Optional suffixes\n"
154            "    'k' or 'K' (kilobyte, 1024), 'M' (megabyte, 1024k), 'G' (gigabyte, 1024M),\n"
155            "    'T' (terabyte, 1024G), 'P' (petabyte, 1024T) and 'E' (exabyte, 1024P)  are\n"
156            "    supported. 'b' is ignored.\n"
157            "  'output_filename' is the destination disk image filename\n"
158            "  'output_fmt' is the destination format\n"
159            "  'options' is a comma separated list of format specific options in a\n"
160            "    name=value format. Use -o ? for an overview of the options supported by the\n"
161            "    used format\n"
162            "  'snapshot_param' is param used for internal snapshot, format\n"
163            "    is 'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
164            "    '[ID_OR_NAME]'\n"
165            "  '-c' indicates that target image must be compressed (qcow format only)\n"
166            "  '-u' allows unsafe backing chains. For rebasing, it is assumed that old and\n"
167            "       new backing file match exactly. The image doesn't need a working\n"
168            "       backing file before rebasing in this case (useful for renaming the\n"
169            "       backing file). For image creation, allow creating without attempting\n"
170            "       to open the backing file.\n"
171            "  '-h' with or without a command shows this help and lists the supported formats\n"
172            "  '-p' show progress of command (only certain commands)\n"
173            "  '-q' use Quiet mode - do not print any output (except errors)\n"
174            "  '-S' indicates the consecutive number of bytes (defaults to 4k) that must\n"
175            "       contain only zeros for qemu-img to create a sparse image during\n"
176            "       conversion. If the number of bytes is 0, the source will not be scanned for\n"
177            "       unallocated or zero sectors, and the destination image will always be\n"
178            "       fully allocated\n"
179            "  '--output' takes the format in which the output must be done (human or json)\n"
180            "  '-n' skips the target volume creation (useful if the volume is created\n"
181            "       prior to running qemu-img)\n"
182            "\n"
183            "Parameters to bitmap subcommand:\n"
184            "  'bitmap' is the name of the bitmap to manipulate, through one or more\n"
185            "       actions from '--add', '--remove', '--clear', '--enable', '--disable',\n"
186            "       or '--merge source'\n"
187            "  '-g granularity' sets the granularity for '--add' actions\n"
188            "  '-b source' and '-F src_fmt' tell '--merge' actions to find the source\n"
189            "       bitmaps from an alternative file\n"
190            "\n"
191            "Parameters to check subcommand:\n"
192            "  '-r' tries to repair any inconsistencies that are found during the check.\n"
193            "       '-r leaks' repairs only cluster leaks, whereas '-r all' fixes all\n"
194            "       kinds of errors, with a higher risk of choosing the wrong fix or\n"
195            "       hiding corruption that has already occurred.\n"
196            "\n"
197            "Parameters to convert subcommand:\n"
198            "  '--bitmaps' copies all top-level persistent bitmaps to destination\n"
199            "  '-m' specifies how many coroutines work in parallel during the convert\n"
200            "       process (defaults to 8)\n"
201            "  '-W' allow to write to the target out of order rather than sequential\n"
202            "\n"
203            "Parameters to snapshot subcommand:\n"
204            "  'snapshot' is the name of the snapshot to create, apply or delete\n"
205            "  '-a' applies a snapshot (revert disk to saved state)\n"
206            "  '-c' creates a snapshot\n"
207            "  '-d' deletes a snapshot\n"
208            "  '-l' lists all snapshots in the given image\n"
209            "\n"
210            "Parameters to compare subcommand:\n"
211            "  '-f' first image format\n"
212            "  '-F' second image format\n"
213            "  '-s' run in Strict mode - fail on different image size or sector allocation\n"
214            "\n"
215            "Parameters to dd subcommand:\n"
216            "  'bs=BYTES' read and write up to BYTES bytes at a time "
217            "(default: 512)\n"
218            "  'count=N' copy only N input blocks\n"
219            "  'if=FILE' read from FILE\n"
220            "  'of=FILE' write to FILE\n"
221            "  'skip=N' skip N bs-sized blocks at the start of input\n";
222 
223     printf("%s\nSupported formats:", help_msg);
224     bdrv_iterate_format(format_print, NULL, false);
225     printf("\n\n" QEMU_HELP_BOTTOM "\n");
226     exit(EXIT_SUCCESS);
227 }
228 
229 /*
230  * Is @optarg safe for accumulate_options()?
231  * It is when multiple of them can be joined together separated by ','.
232  * To make that work, @optarg must not start with ',' (or else a
233  * separating ',' preceding it gets escaped), and it must not end with
234  * an odd number of ',' (or else a separating ',' following it gets
235  * escaped), or be empty (or else a separating ',' preceding it can
236  * escape a separating ',' following it).
237  *
238  */
239 static bool is_valid_option_list(const char *optarg)
240 {
241     size_t len = strlen(optarg);
242     size_t i;
243 
244     if (!optarg[0] || optarg[0] == ',') {
245         return false;
246     }
247 
248     for (i = len; i > 0 && optarg[i - 1] == ','; i--) {
249     }
250     if ((len - i) % 2) {
251         return false;
252     }
253 
254     return true;
255 }
256 
257 static int accumulate_options(char **options, char *optarg)
258 {
259     char *new_options;
260 
261     if (!is_valid_option_list(optarg)) {
262         error_report("Invalid option list: %s", optarg);
263         return -1;
264     }
265 
266     if (!*options) {
267         *options = g_strdup(optarg);
268     } else {
269         new_options = g_strdup_printf("%s,%s", *options, optarg);
270         g_free(*options);
271         *options = new_options;
272     }
273     return 0;
274 }
275 
276 static QemuOptsList qemu_source_opts = {
277     .name = "source",
278     .implied_opt_name = "file",
279     .head = QTAILQ_HEAD_INITIALIZER(qemu_source_opts.head),
280     .desc = {
281         { }
282     },
283 };
284 
285 static int GCC_FMT_ATTR(2, 3) qprintf(bool quiet, const char *fmt, ...)
286 {
287     int ret = 0;
288     if (!quiet) {
289         va_list args;
290         va_start(args, fmt);
291         ret = vprintf(fmt, args);
292         va_end(args);
293     }
294     return ret;
295 }
296 
297 
298 static int print_block_option_help(const char *filename, const char *fmt)
299 {
300     BlockDriver *drv, *proto_drv;
301     QemuOptsList *create_opts = NULL;
302     Error *local_err = NULL;
303 
304     /* Find driver and parse its options */
305     drv = bdrv_find_format(fmt);
306     if (!drv) {
307         error_report("Unknown file format '%s'", fmt);
308         return 1;
309     }
310 
311     if (!drv->create_opts) {
312         error_report("Format driver '%s' does not support image creation", fmt);
313         return 1;
314     }
315 
316     create_opts = qemu_opts_append(create_opts, drv->create_opts);
317     if (filename) {
318         proto_drv = bdrv_find_protocol(filename, true, &local_err);
319         if (!proto_drv) {
320             error_report_err(local_err);
321             qemu_opts_free(create_opts);
322             return 1;
323         }
324         if (!proto_drv->create_opts) {
325             error_report("Protocol driver '%s' does not support image creation",
326                          proto_drv->format_name);
327             qemu_opts_free(create_opts);
328             return 1;
329         }
330         create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
331     }
332 
333     if (filename) {
334         printf("Supported options:\n");
335     } else {
336         printf("Supported %s options:\n", fmt);
337     }
338     qemu_opts_print_help(create_opts, false);
339     qemu_opts_free(create_opts);
340 
341     if (!filename) {
342         printf("\n"
343                "The protocol level may support further options.\n"
344                "Specify the target filename to include those options.\n");
345     }
346 
347     return 0;
348 }
349 
350 
351 static BlockBackend *img_open_opts(const char *optstr,
352                                    QemuOpts *opts, int flags, bool writethrough,
353                                    bool quiet, bool force_share)
354 {
355     QDict *options;
356     Error *local_err = NULL;
357     BlockBackend *blk;
358     options = qemu_opts_to_qdict(opts, NULL);
359     if (force_share) {
360         if (qdict_haskey(options, BDRV_OPT_FORCE_SHARE)
361             && strcmp(qdict_get_str(options, BDRV_OPT_FORCE_SHARE), "on")) {
362             error_report("--force-share/-U conflicts with image options");
363             qobject_unref(options);
364             return NULL;
365         }
366         qdict_put_str(options, BDRV_OPT_FORCE_SHARE, "on");
367     }
368     blk = blk_new_open(NULL, NULL, options, flags, &local_err);
369     if (!blk) {
370         error_reportf_err(local_err, "Could not open '%s': ", optstr);
371         return NULL;
372     }
373     blk_set_enable_write_cache(blk, !writethrough);
374 
375     return blk;
376 }
377 
378 static BlockBackend *img_open_file(const char *filename,
379                                    QDict *options,
380                                    const char *fmt, int flags,
381                                    bool writethrough, bool quiet,
382                                    bool force_share)
383 {
384     BlockBackend *blk;
385     Error *local_err = NULL;
386 
387     if (!options) {
388         options = qdict_new();
389     }
390     if (fmt) {
391         qdict_put_str(options, "driver", fmt);
392     }
393 
394     if (force_share) {
395         qdict_put_bool(options, BDRV_OPT_FORCE_SHARE, true);
396     }
397     blk = blk_new_open(filename, NULL, options, flags, &local_err);
398     if (!blk) {
399         error_reportf_err(local_err, "Could not open '%s': ", filename);
400         return NULL;
401     }
402     blk_set_enable_write_cache(blk, !writethrough);
403 
404     return blk;
405 }
406 
407 
408 static int img_add_key_secrets(void *opaque,
409                                const char *name, const char *value,
410                                Error **errp)
411 {
412     QDict *options = opaque;
413 
414     if (g_str_has_suffix(name, "key-secret")) {
415         qdict_put_str(options, name, value);
416     }
417 
418     return 0;
419 }
420 
421 
422 static BlockBackend *img_open(bool image_opts,
423                               const char *filename,
424                               const char *fmt, int flags, bool writethrough,
425                               bool quiet, bool force_share)
426 {
427     BlockBackend *blk;
428     if (image_opts) {
429         QemuOpts *opts;
430         if (fmt) {
431             error_report("--image-opts and --format are mutually exclusive");
432             return NULL;
433         }
434         opts = qemu_opts_parse_noisily(qemu_find_opts("source"),
435                                        filename, true);
436         if (!opts) {
437             return NULL;
438         }
439         blk = img_open_opts(filename, opts, flags, writethrough, quiet,
440                             force_share);
441     } else {
442         blk = img_open_file(filename, NULL, fmt, flags, writethrough, quiet,
443                             force_share);
444     }
445     return blk;
446 }
447 
448 
449 static int add_old_style_options(const char *fmt, QemuOpts *opts,
450                                  const char *base_filename,
451                                  const char *base_fmt)
452 {
453     if (base_filename) {
454         if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
455                           NULL)) {
456             error_report("Backing file not supported for file format '%s'",
457                          fmt);
458             return -1;
459         }
460     }
461     if (base_fmt) {
462         if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
463             error_report("Backing file format not supported for file "
464                          "format '%s'", fmt);
465             return -1;
466         }
467     }
468     return 0;
469 }
470 
471 static int64_t cvtnum_full(const char *name, const char *value, int64_t min,
472                            int64_t max)
473 {
474     int err;
475     uint64_t res;
476 
477     err = qemu_strtosz(value, NULL, &res);
478     if (err < 0 && err != -ERANGE) {
479         error_report("Invalid %s specified. You may use "
480                      "k, M, G, T, P or E suffixes for", name);
481         error_report("kilobytes, megabytes, gigabytes, terabytes, "
482                      "petabytes and exabytes.");
483         return err;
484     }
485     if (err == -ERANGE || res > max || res < min) {
486         error_report("Invalid %s specified. Must be between %" PRId64
487                      " and %" PRId64 ".", name, min, max);
488         return -ERANGE;
489     }
490     return res;
491 }
492 
493 static int64_t cvtnum(const char *name, const char *value)
494 {
495     return cvtnum_full(name, value, 0, INT64_MAX);
496 }
497 
498 static int img_create(int argc, char **argv)
499 {
500     int c;
501     uint64_t img_size = -1;
502     const char *fmt = "raw";
503     const char *base_fmt = NULL;
504     const char *filename;
505     const char *base_filename = NULL;
506     char *options = NULL;
507     Error *local_err = NULL;
508     bool quiet = false;
509     int flags = 0;
510 
511     for(;;) {
512         static const struct option long_options[] = {
513             {"help", no_argument, 0, 'h'},
514             {"object", required_argument, 0, OPTION_OBJECT},
515             {0, 0, 0, 0}
516         };
517         c = getopt_long(argc, argv, ":F:b:f:ho:qu",
518                         long_options, NULL);
519         if (c == -1) {
520             break;
521         }
522         switch(c) {
523         case ':':
524             missing_argument(argv[optind - 1]);
525             break;
526         case '?':
527             unrecognized_option(argv[optind - 1]);
528             break;
529         case 'h':
530             help();
531             break;
532         case 'F':
533             base_fmt = optarg;
534             break;
535         case 'b':
536             base_filename = optarg;
537             break;
538         case 'f':
539             fmt = optarg;
540             break;
541         case 'o':
542             if (accumulate_options(&options, optarg) < 0) {
543                 goto fail;
544             }
545             break;
546         case 'q':
547             quiet = true;
548             break;
549         case 'u':
550             flags |= BDRV_O_NO_BACKING;
551             break;
552         case OPTION_OBJECT:
553             user_creatable_process_cmdline(optarg);
554             break;
555         }
556     }
557 
558     /* Get the filename */
559     filename = (optind < argc) ? argv[optind] : NULL;
560     if (options && has_help_option(options)) {
561         g_free(options);
562         return print_block_option_help(filename, fmt);
563     }
564 
565     if (optind >= argc) {
566         error_exit("Expecting image file name");
567     }
568     optind++;
569 
570     /* Get image size, if specified */
571     if (optind < argc) {
572         int64_t sval;
573 
574         sval = cvtnum("image size", argv[optind++]);
575         if (sval < 0) {
576             goto fail;
577         }
578         img_size = (uint64_t)sval;
579     }
580     if (optind != argc) {
581         error_exit("Unexpected argument: %s", argv[optind]);
582     }
583 
584     bdrv_img_create(filename, fmt, base_filename, base_fmt,
585                     options, img_size, flags, quiet, &local_err);
586     if (local_err) {
587         error_reportf_err(local_err, "%s: ", filename);
588         goto fail;
589     }
590 
591     g_free(options);
592     return 0;
593 
594 fail:
595     g_free(options);
596     return 1;
597 }
598 
599 static void dump_json_image_check(ImageCheck *check, bool quiet)
600 {
601     GString *str;
602     QObject *obj;
603     Visitor *v = qobject_output_visitor_new(&obj);
604 
605     visit_type_ImageCheck(v, NULL, &check, &error_abort);
606     visit_complete(v, &obj);
607     str = qobject_to_json_pretty(obj, true);
608     assert(str != NULL);
609     qprintf(quiet, "%s\n", str->str);
610     qobject_unref(obj);
611     visit_free(v);
612     g_string_free(str, true);
613 }
614 
615 static void dump_human_image_check(ImageCheck *check, bool quiet)
616 {
617     if (!(check->corruptions || check->leaks || check->check_errors)) {
618         qprintf(quiet, "No errors were found on the image.\n");
619     } else {
620         if (check->corruptions) {
621             qprintf(quiet, "\n%" PRId64 " errors were found on the image.\n"
622                     "Data may be corrupted, or further writes to the image "
623                     "may corrupt it.\n",
624                     check->corruptions);
625         }
626 
627         if (check->leaks) {
628             qprintf(quiet,
629                     "\n%" PRId64 " leaked clusters were found on the image.\n"
630                     "This means waste of disk space, but no harm to data.\n",
631                     check->leaks);
632         }
633 
634         if (check->check_errors) {
635             qprintf(quiet,
636                     "\n%" PRId64
637                     " internal errors have occurred during the check.\n",
638                     check->check_errors);
639         }
640     }
641 
642     if (check->total_clusters != 0 && check->allocated_clusters != 0) {
643         qprintf(quiet, "%" PRId64 "/%" PRId64 " = %0.2f%% allocated, "
644                 "%0.2f%% fragmented, %0.2f%% compressed clusters\n",
645                 check->allocated_clusters, check->total_clusters,
646                 check->allocated_clusters * 100.0 / check->total_clusters,
647                 check->fragmented_clusters * 100.0 / check->allocated_clusters,
648                 check->compressed_clusters * 100.0 /
649                 check->allocated_clusters);
650     }
651 
652     if (check->image_end_offset) {
653         qprintf(quiet,
654                 "Image end offset: %" PRId64 "\n", check->image_end_offset);
655     }
656 }
657 
658 static int collect_image_check(BlockDriverState *bs,
659                    ImageCheck *check,
660                    const char *filename,
661                    const char *fmt,
662                    int fix)
663 {
664     int ret;
665     BdrvCheckResult result;
666 
667     ret = bdrv_check(bs, &result, fix);
668     if (ret < 0) {
669         return ret;
670     }
671 
672     check->filename                 = g_strdup(filename);
673     check->format                   = g_strdup(bdrv_get_format_name(bs));
674     check->check_errors             = result.check_errors;
675     check->corruptions              = result.corruptions;
676     check->has_corruptions          = result.corruptions != 0;
677     check->leaks                    = result.leaks;
678     check->has_leaks                = result.leaks != 0;
679     check->corruptions_fixed        = result.corruptions_fixed;
680     check->has_corruptions_fixed    = result.corruptions_fixed != 0;
681     check->leaks_fixed              = result.leaks_fixed;
682     check->has_leaks_fixed          = result.leaks_fixed != 0;
683     check->image_end_offset         = result.image_end_offset;
684     check->has_image_end_offset     = result.image_end_offset != 0;
685     check->total_clusters           = result.bfi.total_clusters;
686     check->has_total_clusters       = result.bfi.total_clusters != 0;
687     check->allocated_clusters       = result.bfi.allocated_clusters;
688     check->has_allocated_clusters   = result.bfi.allocated_clusters != 0;
689     check->fragmented_clusters      = result.bfi.fragmented_clusters;
690     check->has_fragmented_clusters  = result.bfi.fragmented_clusters != 0;
691     check->compressed_clusters      = result.bfi.compressed_clusters;
692     check->has_compressed_clusters  = result.bfi.compressed_clusters != 0;
693 
694     return 0;
695 }
696 
697 /*
698  * Checks an image for consistency. Exit codes:
699  *
700  *  0 - Check completed, image is good
701  *  1 - Check not completed because of internal errors
702  *  2 - Check completed, image is corrupted
703  *  3 - Check completed, image has leaked clusters, but is good otherwise
704  * 63 - Checks are not supported by the image format
705  */
706 static int img_check(int argc, char **argv)
707 {
708     int c, ret;
709     OutputFormat output_format = OFORMAT_HUMAN;
710     const char *filename, *fmt, *output, *cache;
711     BlockBackend *blk;
712     BlockDriverState *bs;
713     int fix = 0;
714     int flags = BDRV_O_CHECK;
715     bool writethrough;
716     ImageCheck *check;
717     bool quiet = false;
718     bool image_opts = false;
719     bool force_share = false;
720 
721     fmt = NULL;
722     output = NULL;
723     cache = BDRV_DEFAULT_CACHE;
724 
725     for(;;) {
726         int option_index = 0;
727         static const struct option long_options[] = {
728             {"help", no_argument, 0, 'h'},
729             {"format", required_argument, 0, 'f'},
730             {"repair", required_argument, 0, 'r'},
731             {"output", required_argument, 0, OPTION_OUTPUT},
732             {"object", required_argument, 0, OPTION_OBJECT},
733             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
734             {"force-share", no_argument, 0, 'U'},
735             {0, 0, 0, 0}
736         };
737         c = getopt_long(argc, argv, ":hf:r:T:qU",
738                         long_options, &option_index);
739         if (c == -1) {
740             break;
741         }
742         switch(c) {
743         case ':':
744             missing_argument(argv[optind - 1]);
745             break;
746         case '?':
747             unrecognized_option(argv[optind - 1]);
748             break;
749         case 'h':
750             help();
751             break;
752         case 'f':
753             fmt = optarg;
754             break;
755         case 'r':
756             flags |= BDRV_O_RDWR;
757 
758             if (!strcmp(optarg, "leaks")) {
759                 fix = BDRV_FIX_LEAKS;
760             } else if (!strcmp(optarg, "all")) {
761                 fix = BDRV_FIX_LEAKS | BDRV_FIX_ERRORS;
762             } else {
763                 error_exit("Unknown option value for -r "
764                            "(expecting 'leaks' or 'all'): %s", optarg);
765             }
766             break;
767         case OPTION_OUTPUT:
768             output = optarg;
769             break;
770         case 'T':
771             cache = optarg;
772             break;
773         case 'q':
774             quiet = true;
775             break;
776         case 'U':
777             force_share = true;
778             break;
779         case OPTION_OBJECT:
780             user_creatable_process_cmdline(optarg);
781             break;
782         case OPTION_IMAGE_OPTS:
783             image_opts = true;
784             break;
785         }
786     }
787     if (optind != argc - 1) {
788         error_exit("Expecting one image file name");
789     }
790     filename = argv[optind++];
791 
792     if (output && !strcmp(output, "json")) {
793         output_format = OFORMAT_JSON;
794     } else if (output && !strcmp(output, "human")) {
795         output_format = OFORMAT_HUMAN;
796     } else if (output) {
797         error_report("--output must be used with human or json as argument.");
798         return 1;
799     }
800 
801     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
802     if (ret < 0) {
803         error_report("Invalid source cache option: %s", cache);
804         return 1;
805     }
806 
807     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
808                    force_share);
809     if (!blk) {
810         return 1;
811     }
812     bs = blk_bs(blk);
813 
814     check = g_new0(ImageCheck, 1);
815     ret = collect_image_check(bs, check, filename, fmt, fix);
816 
817     if (ret == -ENOTSUP) {
818         error_report("This image format does not support checks");
819         ret = 63;
820         goto fail;
821     }
822 
823     if (check->corruptions_fixed || check->leaks_fixed) {
824         int corruptions_fixed, leaks_fixed;
825         bool has_leaks_fixed, has_corruptions_fixed;
826 
827         leaks_fixed         = check->leaks_fixed;
828         has_leaks_fixed     = check->has_leaks_fixed;
829         corruptions_fixed   = check->corruptions_fixed;
830         has_corruptions_fixed = check->has_corruptions_fixed;
831 
832         if (output_format == OFORMAT_HUMAN) {
833             qprintf(quiet,
834                     "The following inconsistencies were found and repaired:\n\n"
835                     "    %" PRId64 " leaked clusters\n"
836                     "    %" PRId64 " corruptions\n\n"
837                     "Double checking the fixed image now...\n",
838                     check->leaks_fixed,
839                     check->corruptions_fixed);
840         }
841 
842         qapi_free_ImageCheck(check);
843         check = g_new0(ImageCheck, 1);
844         ret = collect_image_check(bs, check, filename, fmt, 0);
845 
846         check->leaks_fixed          = leaks_fixed;
847         check->has_leaks_fixed      = has_leaks_fixed;
848         check->corruptions_fixed    = corruptions_fixed;
849         check->has_corruptions_fixed = has_corruptions_fixed;
850     }
851 
852     if (!ret) {
853         switch (output_format) {
854         case OFORMAT_HUMAN:
855             dump_human_image_check(check, quiet);
856             break;
857         case OFORMAT_JSON:
858             dump_json_image_check(check, quiet);
859             break;
860         }
861     }
862 
863     if (ret || check->check_errors) {
864         if (ret) {
865             error_report("Check failed: %s", strerror(-ret));
866         } else {
867             error_report("Check failed");
868         }
869         ret = 1;
870         goto fail;
871     }
872 
873     if (check->corruptions) {
874         ret = 2;
875     } else if (check->leaks) {
876         ret = 3;
877     } else {
878         ret = 0;
879     }
880 
881 fail:
882     qapi_free_ImageCheck(check);
883     blk_unref(blk);
884     return ret;
885 }
886 
887 typedef struct CommonBlockJobCBInfo {
888     BlockDriverState *bs;
889     Error **errp;
890 } CommonBlockJobCBInfo;
891 
892 static void common_block_job_cb(void *opaque, int ret)
893 {
894     CommonBlockJobCBInfo *cbi = opaque;
895 
896     if (ret < 0) {
897         error_setg_errno(cbi->errp, -ret, "Block job failed");
898     }
899 }
900 
901 static void run_block_job(BlockJob *job, Error **errp)
902 {
903     AioContext *aio_context = blk_get_aio_context(job->blk);
904     int ret = 0;
905 
906     aio_context_acquire(aio_context);
907     job_ref(&job->job);
908     do {
909         float progress = 0.0f;
910         aio_poll(aio_context, true);
911         if (job->job.progress.total) {
912             progress = (float)job->job.progress.current /
913                        job->job.progress.total * 100.f;
914         }
915         qemu_progress_print(progress, 0);
916     } while (!job_is_ready(&job->job) && !job_is_completed(&job->job));
917 
918     if (!job_is_completed(&job->job)) {
919         ret = job_complete_sync(&job->job, errp);
920     } else {
921         ret = job->job.ret;
922     }
923     job_unref(&job->job);
924     aio_context_release(aio_context);
925 
926     /* publish completion progress only when success */
927     if (!ret) {
928         qemu_progress_print(100.f, 0);
929     }
930 }
931 
932 static int img_commit(int argc, char **argv)
933 {
934     int c, ret, flags;
935     const char *filename, *fmt, *cache, *base;
936     BlockBackend *blk;
937     BlockDriverState *bs, *base_bs;
938     BlockJob *job;
939     bool progress = false, quiet = false, drop = false;
940     bool writethrough;
941     Error *local_err = NULL;
942     CommonBlockJobCBInfo cbi;
943     bool image_opts = false;
944     AioContext *aio_context;
945     int64_t rate_limit = 0;
946 
947     fmt = NULL;
948     cache = BDRV_DEFAULT_CACHE;
949     base = NULL;
950     for(;;) {
951         static const struct option long_options[] = {
952             {"help", no_argument, 0, 'h'},
953             {"object", required_argument, 0, OPTION_OBJECT},
954             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
955             {0, 0, 0, 0}
956         };
957         c = getopt_long(argc, argv, ":f:ht:b:dpqr:",
958                         long_options, NULL);
959         if (c == -1) {
960             break;
961         }
962         switch(c) {
963         case ':':
964             missing_argument(argv[optind - 1]);
965             break;
966         case '?':
967             unrecognized_option(argv[optind - 1]);
968             break;
969         case 'h':
970             help();
971             break;
972         case 'f':
973             fmt = optarg;
974             break;
975         case 't':
976             cache = optarg;
977             break;
978         case 'b':
979             base = optarg;
980             /* -b implies -d */
981             drop = true;
982             break;
983         case 'd':
984             drop = true;
985             break;
986         case 'p':
987             progress = true;
988             break;
989         case 'q':
990             quiet = true;
991             break;
992         case 'r':
993             rate_limit = cvtnum("rate limit", optarg);
994             if (rate_limit < 0) {
995                 return 1;
996             }
997             break;
998         case OPTION_OBJECT:
999             user_creatable_process_cmdline(optarg);
1000             break;
1001         case OPTION_IMAGE_OPTS:
1002             image_opts = true;
1003             break;
1004         }
1005     }
1006 
1007     /* Progress is not shown in Quiet mode */
1008     if (quiet) {
1009         progress = false;
1010     }
1011 
1012     if (optind != argc - 1) {
1013         error_exit("Expecting one image file name");
1014     }
1015     filename = argv[optind++];
1016 
1017     flags = BDRV_O_RDWR | BDRV_O_UNMAP;
1018     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
1019     if (ret < 0) {
1020         error_report("Invalid cache option: %s", cache);
1021         return 1;
1022     }
1023 
1024     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
1025                    false);
1026     if (!blk) {
1027         return 1;
1028     }
1029     bs = blk_bs(blk);
1030 
1031     qemu_progress_init(progress, 1.f);
1032     qemu_progress_print(0.f, 100);
1033 
1034     if (base) {
1035         base_bs = bdrv_find_backing_image(bs, base);
1036         if (!base_bs) {
1037             error_setg(&local_err,
1038                        "Did not find '%s' in the backing chain of '%s'",
1039                        base, filename);
1040             goto done;
1041         }
1042     } else {
1043         /* This is different from QMP, which by default uses the deepest file in
1044          * the backing chain (i.e., the very base); however, the traditional
1045          * behavior of qemu-img commit is using the immediate backing file. */
1046         base_bs = bdrv_backing_chain_next(bs);
1047         if (!base_bs) {
1048             error_setg(&local_err, "Image does not have a backing file");
1049             goto done;
1050         }
1051     }
1052 
1053     cbi = (CommonBlockJobCBInfo){
1054         .errp = &local_err,
1055         .bs   = bs,
1056     };
1057 
1058     aio_context = bdrv_get_aio_context(bs);
1059     aio_context_acquire(aio_context);
1060     commit_active_start("commit", bs, base_bs, JOB_DEFAULT, rate_limit,
1061                         BLOCKDEV_ON_ERROR_REPORT, NULL, common_block_job_cb,
1062                         &cbi, false, &local_err);
1063     aio_context_release(aio_context);
1064     if (local_err) {
1065         goto done;
1066     }
1067 
1068     /* When the block job completes, the BlockBackend reference will point to
1069      * the old backing file. In order to avoid that the top image is already
1070      * deleted, so we can still empty it afterwards, increment the reference
1071      * counter here preemptively. */
1072     if (!drop) {
1073         bdrv_ref(bs);
1074     }
1075 
1076     job = block_job_get("commit");
1077     assert(job);
1078     run_block_job(job, &local_err);
1079     if (local_err) {
1080         goto unref_backing;
1081     }
1082 
1083     if (!drop) {
1084         BlockBackend *old_backing_blk;
1085 
1086         old_backing_blk = blk_new_with_bs(bs, BLK_PERM_WRITE, BLK_PERM_ALL,
1087                                           &local_err);
1088         if (!old_backing_blk) {
1089             goto unref_backing;
1090         }
1091         ret = blk_make_empty(old_backing_blk, &local_err);
1092         blk_unref(old_backing_blk);
1093         if (ret == -ENOTSUP) {
1094             error_free(local_err);
1095             local_err = NULL;
1096         } else if (ret < 0) {
1097             goto unref_backing;
1098         }
1099     }
1100 
1101 unref_backing:
1102     if (!drop) {
1103         bdrv_unref(bs);
1104     }
1105 
1106 done:
1107     qemu_progress_end();
1108 
1109     blk_unref(blk);
1110 
1111     if (local_err) {
1112         error_report_err(local_err);
1113         return 1;
1114     }
1115 
1116     qprintf(quiet, "Image committed.\n");
1117     return 0;
1118 }
1119 
1120 /*
1121  * Returns -1 if 'buf' contains only zeroes, otherwise the byte index
1122  * of the first sector boundary within buf where the sector contains a
1123  * non-zero byte.  This function is robust to a buffer that is not
1124  * sector-aligned.
1125  */
1126 static int64_t find_nonzero(const uint8_t *buf, int64_t n)
1127 {
1128     int64_t i;
1129     int64_t end = QEMU_ALIGN_DOWN(n, BDRV_SECTOR_SIZE);
1130 
1131     for (i = 0; i < end; i += BDRV_SECTOR_SIZE) {
1132         if (!buffer_is_zero(buf + i, BDRV_SECTOR_SIZE)) {
1133             return i;
1134         }
1135     }
1136     if (i < n && !buffer_is_zero(buf + i, n - end)) {
1137         return i;
1138     }
1139     return -1;
1140 }
1141 
1142 /*
1143  * Returns true iff the first sector pointed to by 'buf' contains at least
1144  * a non-NUL byte.
1145  *
1146  * 'pnum' is set to the number of sectors (including and immediately following
1147  * the first one) that are known to be in the same allocated/unallocated state.
1148  * The function will try to align the end offset to alignment boundaries so
1149  * that the request will at least end aligned and consecutive requests will
1150  * also start at an aligned offset.
1151  */
1152 static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum,
1153                                 int64_t sector_num, int alignment)
1154 {
1155     bool is_zero;
1156     int i, tail;
1157 
1158     if (n <= 0) {
1159         *pnum = 0;
1160         return 0;
1161     }
1162     is_zero = buffer_is_zero(buf, BDRV_SECTOR_SIZE);
1163     for(i = 1; i < n; i++) {
1164         buf += BDRV_SECTOR_SIZE;
1165         if (is_zero != buffer_is_zero(buf, BDRV_SECTOR_SIZE)) {
1166             break;
1167         }
1168     }
1169 
1170     tail = (sector_num + i) & (alignment - 1);
1171     if (tail) {
1172         if (is_zero && i <= tail) {
1173             /* treat unallocated areas which only consist
1174              * of a small tail as allocated. */
1175             is_zero = false;
1176         }
1177         if (!is_zero) {
1178             /* align up end offset of allocated areas. */
1179             i += alignment - tail;
1180             i = MIN(i, n);
1181         } else {
1182             /* align down end offset of zero areas. */
1183             i -= tail;
1184         }
1185     }
1186     *pnum = i;
1187     return !is_zero;
1188 }
1189 
1190 /*
1191  * Like is_allocated_sectors, but if the buffer starts with a used sector,
1192  * up to 'min' consecutive sectors containing zeros are ignored. This avoids
1193  * breaking up write requests for only small sparse areas.
1194  */
1195 static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum,
1196     int min, int64_t sector_num, int alignment)
1197 {
1198     int ret;
1199     int num_checked, num_used;
1200 
1201     if (n < min) {
1202         min = n;
1203     }
1204 
1205     ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
1206     if (!ret) {
1207         return ret;
1208     }
1209 
1210     num_used = *pnum;
1211     buf += BDRV_SECTOR_SIZE * *pnum;
1212     n -= *pnum;
1213     sector_num += *pnum;
1214     num_checked = num_used;
1215 
1216     while (n > 0) {
1217         ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
1218 
1219         buf += BDRV_SECTOR_SIZE * *pnum;
1220         n -= *pnum;
1221         sector_num += *pnum;
1222         num_checked += *pnum;
1223         if (ret) {
1224             num_used = num_checked;
1225         } else if (*pnum >= min) {
1226             break;
1227         }
1228     }
1229 
1230     *pnum = num_used;
1231     return 1;
1232 }
1233 
1234 /*
1235  * Compares two buffers sector by sector. Returns 0 if the first
1236  * sector of each buffer matches, non-zero otherwise.
1237  *
1238  * pnum is set to the sector-aligned size of the buffer prefix that
1239  * has the same matching status as the first sector.
1240  */
1241 static int compare_buffers(const uint8_t *buf1, const uint8_t *buf2,
1242                            int64_t bytes, int64_t *pnum)
1243 {
1244     bool res;
1245     int64_t i = MIN(bytes, BDRV_SECTOR_SIZE);
1246 
1247     assert(bytes > 0);
1248 
1249     res = !!memcmp(buf1, buf2, i);
1250     while (i < bytes) {
1251         int64_t len = MIN(bytes - i, BDRV_SECTOR_SIZE);
1252 
1253         if (!!memcmp(buf1 + i, buf2 + i, len) != res) {
1254             break;
1255         }
1256         i += len;
1257     }
1258 
1259     *pnum = i;
1260     return res;
1261 }
1262 
1263 #define IO_BUF_SIZE (2 * MiB)
1264 
1265 /*
1266  * Check if passed sectors are empty (not allocated or contain only 0 bytes)
1267  *
1268  * Intended for use by 'qemu-img compare': Returns 0 in case sectors are
1269  * filled with 0, 1 if sectors contain non-zero data (this is a comparison
1270  * failure), and 4 on error (the exit status for read errors), after emitting
1271  * an error message.
1272  *
1273  * @param blk:  BlockBackend for the image
1274  * @param offset: Starting offset to check
1275  * @param bytes: Number of bytes to check
1276  * @param filename: Name of disk file we are checking (logging purpose)
1277  * @param buffer: Allocated buffer for storing read data
1278  * @param quiet: Flag for quiet mode
1279  */
1280 static int check_empty_sectors(BlockBackend *blk, int64_t offset,
1281                                int64_t bytes, const char *filename,
1282                                uint8_t *buffer, bool quiet)
1283 {
1284     int ret = 0;
1285     int64_t idx;
1286 
1287     ret = blk_pread(blk, offset, buffer, bytes);
1288     if (ret < 0) {
1289         error_report("Error while reading offset %" PRId64 " of %s: %s",
1290                      offset, filename, strerror(-ret));
1291         return 4;
1292     }
1293     idx = find_nonzero(buffer, bytes);
1294     if (idx >= 0) {
1295         qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1296                 offset + idx);
1297         return 1;
1298     }
1299 
1300     return 0;
1301 }
1302 
1303 /*
1304  * Compares two images. Exit codes:
1305  *
1306  * 0 - Images are identical or the requested help was printed
1307  * 1 - Images differ
1308  * >1 - Error occurred
1309  */
1310 static int img_compare(int argc, char **argv)
1311 {
1312     const char *fmt1 = NULL, *fmt2 = NULL, *cache, *filename1, *filename2;
1313     BlockBackend *blk1, *blk2;
1314     BlockDriverState *bs1, *bs2;
1315     int64_t total_size1, total_size2;
1316     uint8_t *buf1 = NULL, *buf2 = NULL;
1317     int64_t pnum1, pnum2;
1318     int allocated1, allocated2;
1319     int ret = 0; /* return value - 0 Ident, 1 Different, >1 Error */
1320     bool progress = false, quiet = false, strict = false;
1321     int flags;
1322     bool writethrough;
1323     int64_t total_size;
1324     int64_t offset = 0;
1325     int64_t chunk;
1326     int c;
1327     uint64_t progress_base;
1328     bool image_opts = false;
1329     bool force_share = false;
1330 
1331     cache = BDRV_DEFAULT_CACHE;
1332     for (;;) {
1333         static const struct option long_options[] = {
1334             {"help", no_argument, 0, 'h'},
1335             {"object", required_argument, 0, OPTION_OBJECT},
1336             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
1337             {"force-share", no_argument, 0, 'U'},
1338             {0, 0, 0, 0}
1339         };
1340         c = getopt_long(argc, argv, ":hf:F:T:pqsU",
1341                         long_options, NULL);
1342         if (c == -1) {
1343             break;
1344         }
1345         switch (c) {
1346         case ':':
1347             missing_argument(argv[optind - 1]);
1348             break;
1349         case '?':
1350             unrecognized_option(argv[optind - 1]);
1351             break;
1352         case 'h':
1353             help();
1354             break;
1355         case 'f':
1356             fmt1 = optarg;
1357             break;
1358         case 'F':
1359             fmt2 = optarg;
1360             break;
1361         case 'T':
1362             cache = optarg;
1363             break;
1364         case 'p':
1365             progress = true;
1366             break;
1367         case 'q':
1368             quiet = true;
1369             break;
1370         case 's':
1371             strict = true;
1372             break;
1373         case 'U':
1374             force_share = true;
1375             break;
1376         case OPTION_OBJECT:
1377             {
1378                 Error *local_err = NULL;
1379 
1380                 if (!user_creatable_add_from_str(optarg, &local_err)) {
1381                     if (local_err) {
1382                         error_report_err(local_err);
1383                         exit(2);
1384                     } else {
1385                         /* Help was printed */
1386                         exit(EXIT_SUCCESS);
1387                     }
1388                 }
1389                 break;
1390             }
1391         case OPTION_IMAGE_OPTS:
1392             image_opts = true;
1393             break;
1394         }
1395     }
1396 
1397     /* Progress is not shown in Quiet mode */
1398     if (quiet) {
1399         progress = false;
1400     }
1401 
1402 
1403     if (optind != argc - 2) {
1404         error_exit("Expecting two image file names");
1405     }
1406     filename1 = argv[optind++];
1407     filename2 = argv[optind++];
1408 
1409     /* Initialize before goto out */
1410     qemu_progress_init(progress, 2.0);
1411 
1412     flags = 0;
1413     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
1414     if (ret < 0) {
1415         error_report("Invalid source cache option: %s", cache);
1416         ret = 2;
1417         goto out3;
1418     }
1419 
1420     blk1 = img_open(image_opts, filename1, fmt1, flags, writethrough, quiet,
1421                     force_share);
1422     if (!blk1) {
1423         ret = 2;
1424         goto out3;
1425     }
1426 
1427     blk2 = img_open(image_opts, filename2, fmt2, flags, writethrough, quiet,
1428                     force_share);
1429     if (!blk2) {
1430         ret = 2;
1431         goto out2;
1432     }
1433     bs1 = blk_bs(blk1);
1434     bs2 = blk_bs(blk2);
1435 
1436     buf1 = blk_blockalign(blk1, IO_BUF_SIZE);
1437     buf2 = blk_blockalign(blk2, IO_BUF_SIZE);
1438     total_size1 = blk_getlength(blk1);
1439     if (total_size1 < 0) {
1440         error_report("Can't get size of %s: %s",
1441                      filename1, strerror(-total_size1));
1442         ret = 4;
1443         goto out;
1444     }
1445     total_size2 = blk_getlength(blk2);
1446     if (total_size2 < 0) {
1447         error_report("Can't get size of %s: %s",
1448                      filename2, strerror(-total_size2));
1449         ret = 4;
1450         goto out;
1451     }
1452     total_size = MIN(total_size1, total_size2);
1453     progress_base = MAX(total_size1, total_size2);
1454 
1455     qemu_progress_print(0, 100);
1456 
1457     if (strict && total_size1 != total_size2) {
1458         ret = 1;
1459         qprintf(quiet, "Strict mode: Image size mismatch!\n");
1460         goto out;
1461     }
1462 
1463     while (offset < total_size) {
1464         int status1, status2;
1465 
1466         status1 = bdrv_block_status_above(bs1, NULL, offset,
1467                                           total_size1 - offset, &pnum1, NULL,
1468                                           NULL);
1469         if (status1 < 0) {
1470             ret = 3;
1471             error_report("Sector allocation test failed for %s", filename1);
1472             goto out;
1473         }
1474         allocated1 = status1 & BDRV_BLOCK_ALLOCATED;
1475 
1476         status2 = bdrv_block_status_above(bs2, NULL, offset,
1477                                           total_size2 - offset, &pnum2, NULL,
1478                                           NULL);
1479         if (status2 < 0) {
1480             ret = 3;
1481             error_report("Sector allocation test failed for %s", filename2);
1482             goto out;
1483         }
1484         allocated2 = status2 & BDRV_BLOCK_ALLOCATED;
1485 
1486         assert(pnum1 && pnum2);
1487         chunk = MIN(pnum1, pnum2);
1488 
1489         if (strict) {
1490             if (status1 != status2) {
1491                 ret = 1;
1492                 qprintf(quiet, "Strict mode: Offset %" PRId64
1493                         " block status mismatch!\n", offset);
1494                 goto out;
1495             }
1496         }
1497         if ((status1 & BDRV_BLOCK_ZERO) && (status2 & BDRV_BLOCK_ZERO)) {
1498             /* nothing to do */
1499         } else if (allocated1 == allocated2) {
1500             if (allocated1) {
1501                 int64_t pnum;
1502 
1503                 chunk = MIN(chunk, IO_BUF_SIZE);
1504                 ret = blk_pread(blk1, offset, buf1, chunk);
1505                 if (ret < 0) {
1506                     error_report("Error while reading offset %" PRId64
1507                                  " of %s: %s",
1508                                  offset, filename1, strerror(-ret));
1509                     ret = 4;
1510                     goto out;
1511                 }
1512                 ret = blk_pread(blk2, offset, buf2, chunk);
1513                 if (ret < 0) {
1514                     error_report("Error while reading offset %" PRId64
1515                                  " of %s: %s",
1516                                  offset, filename2, strerror(-ret));
1517                     ret = 4;
1518                     goto out;
1519                 }
1520                 ret = compare_buffers(buf1, buf2, chunk, &pnum);
1521                 if (ret || pnum != chunk) {
1522                     qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1523                             offset + (ret ? 0 : pnum));
1524                     ret = 1;
1525                     goto out;
1526                 }
1527             }
1528         } else {
1529             chunk = MIN(chunk, IO_BUF_SIZE);
1530             if (allocated1) {
1531                 ret = check_empty_sectors(blk1, offset, chunk,
1532                                           filename1, buf1, quiet);
1533             } else {
1534                 ret = check_empty_sectors(blk2, offset, chunk,
1535                                           filename2, buf1, quiet);
1536             }
1537             if (ret) {
1538                 goto out;
1539             }
1540         }
1541         offset += chunk;
1542         qemu_progress_print(((float) chunk / progress_base) * 100, 100);
1543     }
1544 
1545     if (total_size1 != total_size2) {
1546         BlockBackend *blk_over;
1547         const char *filename_over;
1548 
1549         qprintf(quiet, "Warning: Image size mismatch!\n");
1550         if (total_size1 > total_size2) {
1551             blk_over = blk1;
1552             filename_over = filename1;
1553         } else {
1554             blk_over = blk2;
1555             filename_over = filename2;
1556         }
1557 
1558         while (offset < progress_base) {
1559             ret = bdrv_block_status_above(blk_bs(blk_over), NULL, offset,
1560                                           progress_base - offset, &chunk,
1561                                           NULL, NULL);
1562             if (ret < 0) {
1563                 ret = 3;
1564                 error_report("Sector allocation test failed for %s",
1565                              filename_over);
1566                 goto out;
1567 
1568             }
1569             if (ret & BDRV_BLOCK_ALLOCATED && !(ret & BDRV_BLOCK_ZERO)) {
1570                 chunk = MIN(chunk, IO_BUF_SIZE);
1571                 ret = check_empty_sectors(blk_over, offset, chunk,
1572                                           filename_over, buf1, quiet);
1573                 if (ret) {
1574                     goto out;
1575                 }
1576             }
1577             offset += chunk;
1578             qemu_progress_print(((float) chunk / progress_base) * 100, 100);
1579         }
1580     }
1581 
1582     qprintf(quiet, "Images are identical.\n");
1583     ret = 0;
1584 
1585 out:
1586     qemu_vfree(buf1);
1587     qemu_vfree(buf2);
1588     blk_unref(blk2);
1589 out2:
1590     blk_unref(blk1);
1591 out3:
1592     qemu_progress_end();
1593     return ret;
1594 }
1595 
1596 /* Convenience wrapper around qmp_block_dirty_bitmap_merge */
1597 static void do_dirty_bitmap_merge(const char *dst_node, const char *dst_name,
1598                                   const char *src_node, const char *src_name,
1599                                   Error **errp)
1600 {
1601     BlockDirtyBitmapMergeSource *merge_src;
1602     BlockDirtyBitmapMergeSourceList *list = NULL;
1603 
1604     merge_src = g_new0(BlockDirtyBitmapMergeSource, 1);
1605     merge_src->type = QTYPE_QDICT;
1606     merge_src->u.external.node = g_strdup(src_node);
1607     merge_src->u.external.name = g_strdup(src_name);
1608     QAPI_LIST_PREPEND(list, merge_src);
1609     qmp_block_dirty_bitmap_merge(dst_node, dst_name, list, errp);
1610     qapi_free_BlockDirtyBitmapMergeSourceList(list);
1611 }
1612 
1613 enum ImgConvertBlockStatus {
1614     BLK_DATA,
1615     BLK_ZERO,
1616     BLK_BACKING_FILE,
1617 };
1618 
1619 #define MAX_COROUTINES 16
1620 #define CONVERT_THROTTLE_GROUP "img_convert"
1621 
1622 typedef struct ImgConvertState {
1623     BlockBackend **src;
1624     int64_t *src_sectors;
1625     int *src_alignment;
1626     int src_num;
1627     int64_t total_sectors;
1628     int64_t allocated_sectors;
1629     int64_t allocated_done;
1630     int64_t sector_num;
1631     int64_t wr_offs;
1632     enum ImgConvertBlockStatus status;
1633     int64_t sector_next_status;
1634     BlockBackend *target;
1635     bool has_zero_init;
1636     bool compressed;
1637     bool target_is_new;
1638     bool target_has_backing;
1639     int64_t target_backing_sectors; /* negative if unknown */
1640     bool wr_in_order;
1641     bool copy_range;
1642     bool salvage;
1643     bool quiet;
1644     int min_sparse;
1645     int alignment;
1646     size_t cluster_sectors;
1647     size_t buf_sectors;
1648     long num_coroutines;
1649     int running_coroutines;
1650     Coroutine *co[MAX_COROUTINES];
1651     int64_t wait_sector_num[MAX_COROUTINES];
1652     CoMutex lock;
1653     int ret;
1654 } ImgConvertState;
1655 
1656 static void convert_select_part(ImgConvertState *s, int64_t sector_num,
1657                                 int *src_cur, int64_t *src_cur_offset)
1658 {
1659     *src_cur = 0;
1660     *src_cur_offset = 0;
1661     while (sector_num - *src_cur_offset >= s->src_sectors[*src_cur]) {
1662         *src_cur_offset += s->src_sectors[*src_cur];
1663         (*src_cur)++;
1664         assert(*src_cur < s->src_num);
1665     }
1666 }
1667 
1668 static int convert_iteration_sectors(ImgConvertState *s, int64_t sector_num)
1669 {
1670     int64_t src_cur_offset;
1671     int ret, n, src_cur;
1672     bool post_backing_zero = false;
1673 
1674     convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1675 
1676     assert(s->total_sectors > sector_num);
1677     n = MIN(s->total_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
1678 
1679     if (s->target_backing_sectors >= 0) {
1680         if (sector_num >= s->target_backing_sectors) {
1681             post_backing_zero = true;
1682         } else if (sector_num + n > s->target_backing_sectors) {
1683             /* Split requests around target_backing_sectors (because
1684              * starting from there, zeros are handled differently) */
1685             n = s->target_backing_sectors - sector_num;
1686         }
1687     }
1688 
1689     if (s->sector_next_status <= sector_num) {
1690         uint64_t offset = (sector_num - src_cur_offset) * BDRV_SECTOR_SIZE;
1691         int64_t count;
1692         int tail;
1693         BlockDriverState *src_bs = blk_bs(s->src[src_cur]);
1694         BlockDriverState *base;
1695 
1696         if (s->target_has_backing) {
1697             base = bdrv_cow_bs(bdrv_skip_filters(src_bs));
1698         } else {
1699             base = NULL;
1700         }
1701 
1702         do {
1703             count = n * BDRV_SECTOR_SIZE;
1704 
1705             ret = bdrv_block_status_above(src_bs, base, offset, count, &count,
1706                                           NULL, NULL);
1707 
1708             if (ret < 0) {
1709                 if (s->salvage) {
1710                     if (n == 1) {
1711                         if (!s->quiet) {
1712                             warn_report("error while reading block status at "
1713                                         "offset %" PRIu64 ": %s", offset,
1714                                         strerror(-ret));
1715                         }
1716                         /* Just try to read the data, then */
1717                         ret = BDRV_BLOCK_DATA;
1718                         count = BDRV_SECTOR_SIZE;
1719                     } else {
1720                         /* Retry on a shorter range */
1721                         n = DIV_ROUND_UP(n, 4);
1722                     }
1723                 } else {
1724                     error_report("error while reading block status at offset "
1725                                  "%" PRIu64 ": %s", offset, strerror(-ret));
1726                     return ret;
1727                 }
1728             }
1729         } while (ret < 0);
1730 
1731         n = DIV_ROUND_UP(count, BDRV_SECTOR_SIZE);
1732 
1733         /*
1734          * Avoid that s->sector_next_status becomes unaligned to the source
1735          * request alignment and/or cluster size to avoid unnecessary read
1736          * cycles.
1737          */
1738         tail = (sector_num - src_cur_offset + n) % s->src_alignment[src_cur];
1739         if (n > tail) {
1740             n -= tail;
1741         }
1742 
1743         if (ret & BDRV_BLOCK_ZERO) {
1744             s->status = post_backing_zero ? BLK_BACKING_FILE : BLK_ZERO;
1745         } else if (ret & BDRV_BLOCK_DATA) {
1746             s->status = BLK_DATA;
1747         } else {
1748             s->status = s->target_has_backing ? BLK_BACKING_FILE : BLK_DATA;
1749         }
1750 
1751         s->sector_next_status = sector_num + n;
1752     }
1753 
1754     n = MIN(n, s->sector_next_status - sector_num);
1755     if (s->status == BLK_DATA) {
1756         n = MIN(n, s->buf_sectors);
1757     }
1758 
1759     /* We need to write complete clusters for compressed images, so if an
1760      * unallocated area is shorter than that, we must consider the whole
1761      * cluster allocated. */
1762     if (s->compressed) {
1763         if (n < s->cluster_sectors) {
1764             n = MIN(s->cluster_sectors, s->total_sectors - sector_num);
1765             s->status = BLK_DATA;
1766         } else {
1767             n = QEMU_ALIGN_DOWN(n, s->cluster_sectors);
1768         }
1769     }
1770 
1771     return n;
1772 }
1773 
1774 static int coroutine_fn convert_co_read(ImgConvertState *s, int64_t sector_num,
1775                                         int nb_sectors, uint8_t *buf)
1776 {
1777     uint64_t single_read_until = 0;
1778     int n, ret;
1779 
1780     assert(nb_sectors <= s->buf_sectors);
1781     while (nb_sectors > 0) {
1782         BlockBackend *blk;
1783         int src_cur;
1784         int64_t bs_sectors, src_cur_offset;
1785         uint64_t offset;
1786 
1787         /* In the case of compression with multiple source files, we can get a
1788          * nb_sectors that spreads into the next part. So we must be able to
1789          * read across multiple BDSes for one convert_read() call. */
1790         convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1791         blk = s->src[src_cur];
1792         bs_sectors = s->src_sectors[src_cur];
1793 
1794         offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
1795 
1796         n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
1797         if (single_read_until > offset) {
1798             n = 1;
1799         }
1800 
1801         ret = blk_co_pread(blk, offset, n << BDRV_SECTOR_BITS, buf, 0);
1802         if (ret < 0) {
1803             if (s->salvage) {
1804                 if (n > 1) {
1805                     single_read_until = offset + (n << BDRV_SECTOR_BITS);
1806                     continue;
1807                 } else {
1808                     if (!s->quiet) {
1809                         warn_report("error while reading offset %" PRIu64
1810                                     ": %s", offset, strerror(-ret));
1811                     }
1812                     memset(buf, 0, BDRV_SECTOR_SIZE);
1813                 }
1814             } else {
1815                 return ret;
1816             }
1817         }
1818 
1819         sector_num += n;
1820         nb_sectors -= n;
1821         buf += n * BDRV_SECTOR_SIZE;
1822     }
1823 
1824     return 0;
1825 }
1826 
1827 
1828 static int coroutine_fn convert_co_write(ImgConvertState *s, int64_t sector_num,
1829                                          int nb_sectors, uint8_t *buf,
1830                                          enum ImgConvertBlockStatus status)
1831 {
1832     int ret;
1833 
1834     while (nb_sectors > 0) {
1835         int n = nb_sectors;
1836         BdrvRequestFlags flags = s->compressed ? BDRV_REQ_WRITE_COMPRESSED : 0;
1837 
1838         switch (status) {
1839         case BLK_BACKING_FILE:
1840             /* If we have a backing file, leave clusters unallocated that are
1841              * unallocated in the source image, so that the backing file is
1842              * visible at the respective offset. */
1843             assert(s->target_has_backing);
1844             break;
1845 
1846         case BLK_DATA:
1847             /* If we're told to keep the target fully allocated (-S 0) or there
1848              * is real non-zero data, we must write it. Otherwise we can treat
1849              * it as zero sectors.
1850              * Compressed clusters need to be written as a whole, so in that
1851              * case we can only save the write if the buffer is completely
1852              * zeroed. */
1853             if (!s->min_sparse ||
1854                 (!s->compressed &&
1855                  is_allocated_sectors_min(buf, n, &n, s->min_sparse,
1856                                           sector_num, s->alignment)) ||
1857                 (s->compressed &&
1858                  !buffer_is_zero(buf, n * BDRV_SECTOR_SIZE)))
1859             {
1860                 ret = blk_co_pwrite(s->target, sector_num << BDRV_SECTOR_BITS,
1861                                     n << BDRV_SECTOR_BITS, buf, flags);
1862                 if (ret < 0) {
1863                     return ret;
1864                 }
1865                 break;
1866             }
1867             /* fall-through */
1868 
1869         case BLK_ZERO:
1870             if (s->has_zero_init) {
1871                 assert(!s->target_has_backing);
1872                 break;
1873             }
1874             ret = blk_co_pwrite_zeroes(s->target,
1875                                        sector_num << BDRV_SECTOR_BITS,
1876                                        n << BDRV_SECTOR_BITS,
1877                                        BDRV_REQ_MAY_UNMAP);
1878             if (ret < 0) {
1879                 return ret;
1880             }
1881             break;
1882         }
1883 
1884         sector_num += n;
1885         nb_sectors -= n;
1886         buf += n * BDRV_SECTOR_SIZE;
1887     }
1888 
1889     return 0;
1890 }
1891 
1892 static int coroutine_fn convert_co_copy_range(ImgConvertState *s, int64_t sector_num,
1893                                               int nb_sectors)
1894 {
1895     int n, ret;
1896 
1897     while (nb_sectors > 0) {
1898         BlockBackend *blk;
1899         int src_cur;
1900         int64_t bs_sectors, src_cur_offset;
1901         int64_t offset;
1902 
1903         convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1904         offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
1905         blk = s->src[src_cur];
1906         bs_sectors = s->src_sectors[src_cur];
1907 
1908         n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
1909 
1910         ret = blk_co_copy_range(blk, offset, s->target,
1911                                 sector_num << BDRV_SECTOR_BITS,
1912                                 n << BDRV_SECTOR_BITS, 0, 0);
1913         if (ret < 0) {
1914             return ret;
1915         }
1916 
1917         sector_num += n;
1918         nb_sectors -= n;
1919     }
1920     return 0;
1921 }
1922 
1923 static void coroutine_fn convert_co_do_copy(void *opaque)
1924 {
1925     ImgConvertState *s = opaque;
1926     uint8_t *buf = NULL;
1927     int ret, i;
1928     int index = -1;
1929 
1930     for (i = 0; i < s->num_coroutines; i++) {
1931         if (s->co[i] == qemu_coroutine_self()) {
1932             index = i;
1933             break;
1934         }
1935     }
1936     assert(index >= 0);
1937 
1938     s->running_coroutines++;
1939     buf = blk_blockalign(s->target, s->buf_sectors * BDRV_SECTOR_SIZE);
1940 
1941     while (1) {
1942         int n;
1943         int64_t sector_num;
1944         enum ImgConvertBlockStatus status;
1945         bool copy_range;
1946 
1947         qemu_co_mutex_lock(&s->lock);
1948         if (s->ret != -EINPROGRESS || s->sector_num >= s->total_sectors) {
1949             qemu_co_mutex_unlock(&s->lock);
1950             break;
1951         }
1952         n = convert_iteration_sectors(s, s->sector_num);
1953         if (n < 0) {
1954             qemu_co_mutex_unlock(&s->lock);
1955             s->ret = n;
1956             break;
1957         }
1958         /* save current sector and allocation status to local variables */
1959         sector_num = s->sector_num;
1960         status = s->status;
1961         if (!s->min_sparse && s->status == BLK_ZERO) {
1962             n = MIN(n, s->buf_sectors);
1963         }
1964         /* increment global sector counter so that other coroutines can
1965          * already continue reading beyond this request */
1966         s->sector_num += n;
1967         qemu_co_mutex_unlock(&s->lock);
1968 
1969         if (status == BLK_DATA || (!s->min_sparse && status == BLK_ZERO)) {
1970             s->allocated_done += n;
1971             qemu_progress_print(100.0 * s->allocated_done /
1972                                         s->allocated_sectors, 0);
1973         }
1974 
1975 retry:
1976         copy_range = s->copy_range && s->status == BLK_DATA;
1977         if (status == BLK_DATA && !copy_range) {
1978             ret = convert_co_read(s, sector_num, n, buf);
1979             if (ret < 0) {
1980                 error_report("error while reading at byte %lld: %s",
1981                              sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
1982                 s->ret = ret;
1983             }
1984         } else if (!s->min_sparse && status == BLK_ZERO) {
1985             status = BLK_DATA;
1986             memset(buf, 0x00, n * BDRV_SECTOR_SIZE);
1987         }
1988 
1989         if (s->wr_in_order) {
1990             /* keep writes in order */
1991             while (s->wr_offs != sector_num && s->ret == -EINPROGRESS) {
1992                 s->wait_sector_num[index] = sector_num;
1993                 qemu_coroutine_yield();
1994             }
1995             s->wait_sector_num[index] = -1;
1996         }
1997 
1998         if (s->ret == -EINPROGRESS) {
1999             if (copy_range) {
2000                 ret = convert_co_copy_range(s, sector_num, n);
2001                 if (ret) {
2002                     s->copy_range = false;
2003                     goto retry;
2004                 }
2005             } else {
2006                 ret = convert_co_write(s, sector_num, n, buf, status);
2007             }
2008             if (ret < 0) {
2009                 error_report("error while writing at byte %lld: %s",
2010                              sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
2011                 s->ret = ret;
2012             }
2013         }
2014 
2015         if (s->wr_in_order) {
2016             /* reenter the coroutine that might have waited
2017              * for this write to complete */
2018             s->wr_offs = sector_num + n;
2019             for (i = 0; i < s->num_coroutines; i++) {
2020                 if (s->co[i] && s->wait_sector_num[i] == s->wr_offs) {
2021                     /*
2022                      * A -> B -> A cannot occur because A has
2023                      * s->wait_sector_num[i] == -1 during A -> B.  Therefore
2024                      * B will never enter A during this time window.
2025                      */
2026                     qemu_coroutine_enter(s->co[i]);
2027                     break;
2028                 }
2029             }
2030         }
2031     }
2032 
2033     qemu_vfree(buf);
2034     s->co[index] = NULL;
2035     s->running_coroutines--;
2036     if (!s->running_coroutines && s->ret == -EINPROGRESS) {
2037         /* the convert job finished successfully */
2038         s->ret = 0;
2039     }
2040 }
2041 
2042 static int convert_do_copy(ImgConvertState *s)
2043 {
2044     int ret, i, n;
2045     int64_t sector_num = 0;
2046 
2047     /* Check whether we have zero initialisation or can get it efficiently */
2048     if (!s->has_zero_init && s->target_is_new && s->min_sparse &&
2049         !s->target_has_backing) {
2050         s->has_zero_init = bdrv_has_zero_init(blk_bs(s->target));
2051     }
2052 
2053     /* Allocate buffer for copied data. For compressed images, only one cluster
2054      * can be copied at a time. */
2055     if (s->compressed) {
2056         if (s->cluster_sectors <= 0 || s->cluster_sectors > s->buf_sectors) {
2057             error_report("invalid cluster size");
2058             return -EINVAL;
2059         }
2060         s->buf_sectors = s->cluster_sectors;
2061     }
2062 
2063     while (sector_num < s->total_sectors) {
2064         n = convert_iteration_sectors(s, sector_num);
2065         if (n < 0) {
2066             return n;
2067         }
2068         if (s->status == BLK_DATA || (!s->min_sparse && s->status == BLK_ZERO))
2069         {
2070             s->allocated_sectors += n;
2071         }
2072         sector_num += n;
2073     }
2074 
2075     /* Do the copy */
2076     s->sector_next_status = 0;
2077     s->ret = -EINPROGRESS;
2078 
2079     qemu_co_mutex_init(&s->lock);
2080     for (i = 0; i < s->num_coroutines; i++) {
2081         s->co[i] = qemu_coroutine_create(convert_co_do_copy, s);
2082         s->wait_sector_num[i] = -1;
2083         qemu_coroutine_enter(s->co[i]);
2084     }
2085 
2086     while (s->running_coroutines) {
2087         main_loop_wait(false);
2088     }
2089 
2090     if (s->compressed && !s->ret) {
2091         /* signal EOF to align */
2092         ret = blk_pwrite_compressed(s->target, 0, NULL, 0);
2093         if (ret < 0) {
2094             return ret;
2095         }
2096     }
2097 
2098     return s->ret;
2099 }
2100 
2101 static int convert_copy_bitmaps(BlockDriverState *src, BlockDriverState *dst)
2102 {
2103     BdrvDirtyBitmap *bm;
2104     Error *err = NULL;
2105 
2106     FOR_EACH_DIRTY_BITMAP(src, bm) {
2107         const char *name;
2108 
2109         if (!bdrv_dirty_bitmap_get_persistence(bm)) {
2110             continue;
2111         }
2112         name = bdrv_dirty_bitmap_name(bm);
2113         qmp_block_dirty_bitmap_add(dst->node_name, name,
2114                                    true, bdrv_dirty_bitmap_granularity(bm),
2115                                    true, true,
2116                                    true, !bdrv_dirty_bitmap_enabled(bm),
2117                                    &err);
2118         if (err) {
2119             error_reportf_err(err, "Failed to create bitmap %s: ", name);
2120             return -1;
2121         }
2122 
2123         do_dirty_bitmap_merge(dst->node_name, name, src->node_name, name,
2124                               &err);
2125         if (err) {
2126             error_reportf_err(err, "Failed to populate bitmap %s: ", name);
2127             return -1;
2128         }
2129     }
2130 
2131     return 0;
2132 }
2133 
2134 #define MAX_BUF_SECTORS 32768
2135 
2136 static void set_rate_limit(BlockBackend *blk, int64_t rate_limit)
2137 {
2138     ThrottleConfig cfg;
2139 
2140     throttle_config_init(&cfg);
2141     cfg.buckets[THROTTLE_BPS_WRITE].avg = rate_limit;
2142 
2143     blk_io_limits_enable(blk, CONVERT_THROTTLE_GROUP);
2144     blk_set_io_limits(blk, &cfg);
2145 }
2146 
2147 static int img_convert(int argc, char **argv)
2148 {
2149     int c, bs_i, flags, src_flags = BDRV_O_NO_SHARE;
2150     const char *fmt = NULL, *out_fmt = NULL, *cache = "unsafe",
2151                *src_cache = BDRV_DEFAULT_CACHE, *out_baseimg = NULL,
2152                *out_filename, *out_baseimg_param, *snapshot_name = NULL;
2153     BlockDriver *drv = NULL, *proto_drv = NULL;
2154     BlockDriverInfo bdi;
2155     BlockDriverState *out_bs;
2156     QemuOpts *opts = NULL, *sn_opts = NULL;
2157     QemuOptsList *create_opts = NULL;
2158     QDict *open_opts = NULL;
2159     char *options = NULL;
2160     Error *local_err = NULL;
2161     bool writethrough, src_writethrough, image_opts = false,
2162          skip_create = false, progress = false, tgt_image_opts = false;
2163     int64_t ret = -EINVAL;
2164     bool force_share = false;
2165     bool explict_min_sparse = false;
2166     bool bitmaps = false;
2167     int64_t rate_limit = 0;
2168 
2169     ImgConvertState s = (ImgConvertState) {
2170         /* Need at least 4k of zeros for sparse detection */
2171         .min_sparse         = 8,
2172         .copy_range         = false,
2173         .buf_sectors        = IO_BUF_SIZE / BDRV_SECTOR_SIZE,
2174         .wr_in_order        = true,
2175         .num_coroutines     = 8,
2176     };
2177 
2178     for(;;) {
2179         static const struct option long_options[] = {
2180             {"help", no_argument, 0, 'h'},
2181             {"object", required_argument, 0, OPTION_OBJECT},
2182             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2183             {"force-share", no_argument, 0, 'U'},
2184             {"target-image-opts", no_argument, 0, OPTION_TARGET_IMAGE_OPTS},
2185             {"salvage", no_argument, 0, OPTION_SALVAGE},
2186             {"target-is-zero", no_argument, 0, OPTION_TARGET_IS_ZERO},
2187             {"bitmaps", no_argument, 0, OPTION_BITMAPS},
2188             {0, 0, 0, 0}
2189         };
2190         c = getopt_long(argc, argv, ":hf:O:B:Cco:l:S:pt:T:qnm:WUr:",
2191                         long_options, NULL);
2192         if (c == -1) {
2193             break;
2194         }
2195         switch(c) {
2196         case ':':
2197             missing_argument(argv[optind - 1]);
2198             break;
2199         case '?':
2200             unrecognized_option(argv[optind - 1]);
2201             break;
2202         case 'h':
2203             help();
2204             break;
2205         case 'f':
2206             fmt = optarg;
2207             break;
2208         case 'O':
2209             out_fmt = optarg;
2210             break;
2211         case 'B':
2212             out_baseimg = optarg;
2213             break;
2214         case 'C':
2215             s.copy_range = true;
2216             break;
2217         case 'c':
2218             s.compressed = true;
2219             break;
2220         case 'o':
2221             if (accumulate_options(&options, optarg) < 0) {
2222                 goto fail_getopt;
2223             }
2224             break;
2225         case 'l':
2226             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
2227                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
2228                                                   optarg, false);
2229                 if (!sn_opts) {
2230                     error_report("Failed in parsing snapshot param '%s'",
2231                                  optarg);
2232                     goto fail_getopt;
2233                 }
2234             } else {
2235                 snapshot_name = optarg;
2236             }
2237             break;
2238         case 'S':
2239         {
2240             int64_t sval;
2241 
2242             sval = cvtnum("buffer size for sparse output", optarg);
2243             if (sval < 0) {
2244                 goto fail_getopt;
2245             } else if (!QEMU_IS_ALIGNED(sval, BDRV_SECTOR_SIZE) ||
2246                 sval / BDRV_SECTOR_SIZE > MAX_BUF_SECTORS) {
2247                 error_report("Invalid buffer size for sparse output specified. "
2248                     "Valid sizes are multiples of %llu up to %llu. Select "
2249                     "0 to disable sparse detection (fully allocates output).",
2250                     BDRV_SECTOR_SIZE, MAX_BUF_SECTORS * BDRV_SECTOR_SIZE);
2251                 goto fail_getopt;
2252             }
2253 
2254             s.min_sparse = sval / BDRV_SECTOR_SIZE;
2255             explict_min_sparse = true;
2256             break;
2257         }
2258         case 'p':
2259             progress = true;
2260             break;
2261         case 't':
2262             cache = optarg;
2263             break;
2264         case 'T':
2265             src_cache = optarg;
2266             break;
2267         case 'q':
2268             s.quiet = true;
2269             break;
2270         case 'n':
2271             skip_create = true;
2272             break;
2273         case 'm':
2274             if (qemu_strtol(optarg, NULL, 0, &s.num_coroutines) ||
2275                 s.num_coroutines < 1 || s.num_coroutines > MAX_COROUTINES) {
2276                 error_report("Invalid number of coroutines. Allowed number of"
2277                              " coroutines is between 1 and %d", MAX_COROUTINES);
2278                 goto fail_getopt;
2279             }
2280             break;
2281         case 'W':
2282             s.wr_in_order = false;
2283             break;
2284         case 'U':
2285             force_share = true;
2286             break;
2287         case 'r':
2288             rate_limit = cvtnum("rate limit", optarg);
2289             if (rate_limit < 0) {
2290                 goto fail_getopt;
2291             }
2292             break;
2293         case OPTION_OBJECT:
2294             user_creatable_process_cmdline(optarg);
2295             break;
2296         case OPTION_IMAGE_OPTS:
2297             image_opts = true;
2298             break;
2299         case OPTION_SALVAGE:
2300             s.salvage = true;
2301             break;
2302         case OPTION_TARGET_IMAGE_OPTS:
2303             tgt_image_opts = true;
2304             break;
2305         case OPTION_TARGET_IS_ZERO:
2306             /*
2307              * The user asserting that the target is blank has the
2308              * same effect as the target driver supporting zero
2309              * initialisation.
2310              */
2311             s.has_zero_init = true;
2312             break;
2313         case OPTION_BITMAPS:
2314             bitmaps = true;
2315             break;
2316         }
2317     }
2318 
2319     if (!out_fmt && !tgt_image_opts) {
2320         out_fmt = "raw";
2321     }
2322 
2323     if (s.compressed && s.copy_range) {
2324         error_report("Cannot enable copy offloading when -c is used");
2325         goto fail_getopt;
2326     }
2327 
2328     if (explict_min_sparse && s.copy_range) {
2329         error_report("Cannot enable copy offloading when -S is used");
2330         goto fail_getopt;
2331     }
2332 
2333     if (s.copy_range && s.salvage) {
2334         error_report("Cannot use copy offloading in salvaging mode");
2335         goto fail_getopt;
2336     }
2337 
2338     if (tgt_image_opts && !skip_create) {
2339         error_report("--target-image-opts requires use of -n flag");
2340         goto fail_getopt;
2341     }
2342 
2343     if (skip_create && options) {
2344         error_report("-o has no effect when skipping image creation");
2345         goto fail_getopt;
2346     }
2347 
2348     if (s.has_zero_init && !skip_create) {
2349         error_report("--target-is-zero requires use of -n flag");
2350         goto fail_getopt;
2351     }
2352 
2353     s.src_num = argc - optind - 1;
2354     out_filename = s.src_num >= 1 ? argv[argc - 1] : NULL;
2355 
2356     if (options && has_help_option(options)) {
2357         if (out_fmt) {
2358             ret = print_block_option_help(out_filename, out_fmt);
2359             goto fail_getopt;
2360         } else {
2361             error_report("Option help requires a format be specified");
2362             goto fail_getopt;
2363         }
2364     }
2365 
2366     if (s.src_num < 1) {
2367         error_report("Must specify image file name");
2368         goto fail_getopt;
2369     }
2370 
2371     /* ret is still -EINVAL until here */
2372     ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
2373     if (ret < 0) {
2374         error_report("Invalid source cache option: %s", src_cache);
2375         goto fail_getopt;
2376     }
2377 
2378     /* Initialize before goto out */
2379     if (s.quiet) {
2380         progress = false;
2381     }
2382     qemu_progress_init(progress, 1.0);
2383     qemu_progress_print(0, 100);
2384 
2385     s.src = g_new0(BlockBackend *, s.src_num);
2386     s.src_sectors = g_new(int64_t, s.src_num);
2387     s.src_alignment = g_new(int, s.src_num);
2388 
2389     for (bs_i = 0; bs_i < s.src_num; bs_i++) {
2390         BlockDriverState *src_bs;
2391         s.src[bs_i] = img_open(image_opts, argv[optind + bs_i],
2392                                fmt, src_flags, src_writethrough, s.quiet,
2393                                force_share);
2394         if (!s.src[bs_i]) {
2395             ret = -1;
2396             goto out;
2397         }
2398         s.src_sectors[bs_i] = blk_nb_sectors(s.src[bs_i]);
2399         if (s.src_sectors[bs_i] < 0) {
2400             error_report("Could not get size of %s: %s",
2401                          argv[optind + bs_i], strerror(-s.src_sectors[bs_i]));
2402             ret = -1;
2403             goto out;
2404         }
2405         src_bs = blk_bs(s.src[bs_i]);
2406         s.src_alignment[bs_i] = DIV_ROUND_UP(src_bs->bl.request_alignment,
2407                                              BDRV_SECTOR_SIZE);
2408         if (!bdrv_get_info(src_bs, &bdi)) {
2409             s.src_alignment[bs_i] = MAX(s.src_alignment[bs_i],
2410                                         bdi.cluster_size / BDRV_SECTOR_SIZE);
2411         }
2412         s.total_sectors += s.src_sectors[bs_i];
2413     }
2414 
2415     if (sn_opts) {
2416         bdrv_snapshot_load_tmp(blk_bs(s.src[0]),
2417                                qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
2418                                qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
2419                                &local_err);
2420     } else if (snapshot_name != NULL) {
2421         if (s.src_num > 1) {
2422             error_report("No support for concatenating multiple snapshot");
2423             ret = -1;
2424             goto out;
2425         }
2426 
2427         bdrv_snapshot_load_tmp_by_id_or_name(blk_bs(s.src[0]), snapshot_name,
2428                                              &local_err);
2429     }
2430     if (local_err) {
2431         error_reportf_err(local_err, "Failed to load snapshot: ");
2432         ret = -1;
2433         goto out;
2434     }
2435 
2436     if (!skip_create) {
2437         /* Find driver and parse its options */
2438         drv = bdrv_find_format(out_fmt);
2439         if (!drv) {
2440             error_report("Unknown file format '%s'", out_fmt);
2441             ret = -1;
2442             goto out;
2443         }
2444 
2445         proto_drv = bdrv_find_protocol(out_filename, true, &local_err);
2446         if (!proto_drv) {
2447             error_report_err(local_err);
2448             ret = -1;
2449             goto out;
2450         }
2451 
2452         if (!drv->create_opts) {
2453             error_report("Format driver '%s' does not support image creation",
2454                          drv->format_name);
2455             ret = -1;
2456             goto out;
2457         }
2458 
2459         if (!proto_drv->create_opts) {
2460             error_report("Protocol driver '%s' does not support image creation",
2461                          proto_drv->format_name);
2462             ret = -1;
2463             goto out;
2464         }
2465 
2466         create_opts = qemu_opts_append(create_opts, drv->create_opts);
2467         create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
2468 
2469         opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
2470         if (options) {
2471             if (!qemu_opts_do_parse(opts, options, NULL, &local_err)) {
2472                 error_report_err(local_err);
2473                 ret = -1;
2474                 goto out;
2475             }
2476         }
2477 
2478         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2479                             s.total_sectors * BDRV_SECTOR_SIZE, &error_abort);
2480         ret = add_old_style_options(out_fmt, opts, out_baseimg, NULL);
2481         if (ret < 0) {
2482             goto out;
2483         }
2484     }
2485 
2486     /* Get backing file name if -o backing_file was used */
2487     out_baseimg_param = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
2488     if (out_baseimg_param) {
2489         out_baseimg = out_baseimg_param;
2490     }
2491     s.target_has_backing = (bool) out_baseimg;
2492 
2493     if (s.has_zero_init && s.target_has_backing) {
2494         error_report("Cannot use --target-is-zero when the destination "
2495                      "image has a backing file");
2496         goto out;
2497     }
2498 
2499     if (s.src_num > 1 && out_baseimg) {
2500         error_report("Having a backing file for the target makes no sense when "
2501                      "concatenating multiple input images");
2502         ret = -1;
2503         goto out;
2504     }
2505 
2506     if (out_baseimg_param) {
2507         if (!qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT)) {
2508             warn_report("Deprecated use of backing file without explicit "
2509                         "backing format");
2510         }
2511     }
2512 
2513     /* Check if compression is supported */
2514     if (s.compressed) {
2515         bool encryption =
2516             qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, false);
2517         const char *encryptfmt =
2518             qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT);
2519         const char *preallocation =
2520             qemu_opt_get(opts, BLOCK_OPT_PREALLOC);
2521 
2522         if (drv && !block_driver_can_compress(drv)) {
2523             error_report("Compression not supported for this file format");
2524             ret = -1;
2525             goto out;
2526         }
2527 
2528         if (encryption || encryptfmt) {
2529             error_report("Compression and encryption not supported at "
2530                          "the same time");
2531             ret = -1;
2532             goto out;
2533         }
2534 
2535         if (preallocation
2536             && strcmp(preallocation, "off"))
2537         {
2538             error_report("Compression and preallocation not supported at "
2539                          "the same time");
2540             ret = -1;
2541             goto out;
2542         }
2543     }
2544 
2545     /* Determine if bitmaps need copying */
2546     if (bitmaps) {
2547         if (s.src_num > 1) {
2548             error_report("Copying bitmaps only possible with single source");
2549             ret = -1;
2550             goto out;
2551         }
2552         if (!bdrv_supports_persistent_dirty_bitmap(blk_bs(s.src[0]))) {
2553             error_report("Source lacks bitmap support");
2554             ret = -1;
2555             goto out;
2556         }
2557     }
2558 
2559     /*
2560      * The later open call will need any decryption secrets, and
2561      * bdrv_create() will purge "opts", so extract them now before
2562      * they are lost.
2563      */
2564     if (!skip_create) {
2565         open_opts = qdict_new();
2566         qemu_opt_foreach(opts, img_add_key_secrets, open_opts, &error_abort);
2567 
2568         /* Create the new image */
2569         ret = bdrv_create(drv, out_filename, opts, &local_err);
2570         if (ret < 0) {
2571             error_reportf_err(local_err, "%s: error while converting %s: ",
2572                               out_filename, out_fmt);
2573             goto out;
2574         }
2575     }
2576 
2577     s.target_is_new = !skip_create;
2578 
2579     flags = s.min_sparse ? (BDRV_O_RDWR | BDRV_O_UNMAP) : BDRV_O_RDWR;
2580     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
2581     if (ret < 0) {
2582         error_report("Invalid cache option: %s", cache);
2583         goto out;
2584     }
2585 
2586     if (skip_create) {
2587         s.target = img_open(tgt_image_opts, out_filename, out_fmt,
2588                             flags, writethrough, s.quiet, false);
2589     } else {
2590         /* TODO ultimately we should allow --target-image-opts
2591          * to be used even when -n is not given.
2592          * That has to wait for bdrv_create to be improved
2593          * to allow filenames in option syntax
2594          */
2595         s.target = img_open_file(out_filename, open_opts, out_fmt,
2596                                  flags, writethrough, s.quiet, false);
2597         open_opts = NULL; /* blk_new_open will have freed it */
2598     }
2599     if (!s.target) {
2600         ret = -1;
2601         goto out;
2602     }
2603     out_bs = blk_bs(s.target);
2604 
2605     if (bitmaps && !bdrv_supports_persistent_dirty_bitmap(out_bs)) {
2606         error_report("Format driver '%s' does not support bitmaps",
2607                      out_bs->drv->format_name);
2608         ret = -1;
2609         goto out;
2610     }
2611 
2612     if (s.compressed && !block_driver_can_compress(out_bs->drv)) {
2613         error_report("Compression not supported for this file format");
2614         ret = -1;
2615         goto out;
2616     }
2617 
2618     /* increase bufsectors from the default 4096 (2M) if opt_transfer
2619      * or discard_alignment of the out_bs is greater. Limit to
2620      * MAX_BUF_SECTORS as maximum which is currently 32768 (16MB). */
2621     s.buf_sectors = MIN(MAX_BUF_SECTORS,
2622                         MAX(s.buf_sectors,
2623                             MAX(out_bs->bl.opt_transfer >> BDRV_SECTOR_BITS,
2624                                 out_bs->bl.pdiscard_alignment >>
2625                                 BDRV_SECTOR_BITS)));
2626 
2627     /* try to align the write requests to the destination to avoid unnecessary
2628      * RMW cycles. */
2629     s.alignment = MAX(pow2floor(s.min_sparse),
2630                       DIV_ROUND_UP(out_bs->bl.request_alignment,
2631                                    BDRV_SECTOR_SIZE));
2632     assert(is_power_of_2(s.alignment));
2633 
2634     if (skip_create) {
2635         int64_t output_sectors = blk_nb_sectors(s.target);
2636         if (output_sectors < 0) {
2637             error_report("unable to get output image length: %s",
2638                          strerror(-output_sectors));
2639             ret = -1;
2640             goto out;
2641         } else if (output_sectors < s.total_sectors) {
2642             error_report("output file is smaller than input file");
2643             ret = -1;
2644             goto out;
2645         }
2646     }
2647 
2648     if (s.target_has_backing && s.target_is_new) {
2649         /* Errors are treated as "backing length unknown" (which means
2650          * s.target_backing_sectors has to be negative, which it will
2651          * be automatically).  The backing file length is used only
2652          * for optimizations, so such a case is not fatal. */
2653         s.target_backing_sectors =
2654             bdrv_nb_sectors(bdrv_backing_chain_next(out_bs));
2655     } else {
2656         s.target_backing_sectors = -1;
2657     }
2658 
2659     ret = bdrv_get_info(out_bs, &bdi);
2660     if (ret < 0) {
2661         if (s.compressed) {
2662             error_report("could not get block driver info");
2663             goto out;
2664         }
2665     } else {
2666         s.compressed = s.compressed || bdi.needs_compressed_writes;
2667         s.cluster_sectors = bdi.cluster_size / BDRV_SECTOR_SIZE;
2668     }
2669 
2670     if (rate_limit) {
2671         set_rate_limit(s.target, rate_limit);
2672     }
2673 
2674     ret = convert_do_copy(&s);
2675 
2676     /* Now copy the bitmaps */
2677     if (bitmaps && ret == 0) {
2678         ret = convert_copy_bitmaps(blk_bs(s.src[0]), out_bs);
2679     }
2680 
2681 out:
2682     if (!ret) {
2683         qemu_progress_print(100, 0);
2684     }
2685     qemu_progress_end();
2686     qemu_opts_del(opts);
2687     qemu_opts_free(create_opts);
2688     qobject_unref(open_opts);
2689     blk_unref(s.target);
2690     if (s.src) {
2691         for (bs_i = 0; bs_i < s.src_num; bs_i++) {
2692             blk_unref(s.src[bs_i]);
2693         }
2694         g_free(s.src);
2695     }
2696     g_free(s.src_sectors);
2697     g_free(s.src_alignment);
2698 fail_getopt:
2699     qemu_opts_del(sn_opts);
2700     g_free(options);
2701 
2702     return !!ret;
2703 }
2704 
2705 
2706 static void dump_snapshots(BlockDriverState *bs)
2707 {
2708     QEMUSnapshotInfo *sn_tab, *sn;
2709     int nb_sns, i;
2710 
2711     nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2712     if (nb_sns <= 0)
2713         return;
2714     printf("Snapshot list:\n");
2715     bdrv_snapshot_dump(NULL);
2716     printf("\n");
2717     for(i = 0; i < nb_sns; i++) {
2718         sn = &sn_tab[i];
2719         bdrv_snapshot_dump(sn);
2720         printf("\n");
2721     }
2722     g_free(sn_tab);
2723 }
2724 
2725 static void dump_json_image_info_list(ImageInfoList *list)
2726 {
2727     GString *str;
2728     QObject *obj;
2729     Visitor *v = qobject_output_visitor_new(&obj);
2730 
2731     visit_type_ImageInfoList(v, NULL, &list, &error_abort);
2732     visit_complete(v, &obj);
2733     str = qobject_to_json_pretty(obj, true);
2734     assert(str != NULL);
2735     printf("%s\n", str->str);
2736     qobject_unref(obj);
2737     visit_free(v);
2738     g_string_free(str, true);
2739 }
2740 
2741 static void dump_json_image_info(ImageInfo *info)
2742 {
2743     GString *str;
2744     QObject *obj;
2745     Visitor *v = qobject_output_visitor_new(&obj);
2746 
2747     visit_type_ImageInfo(v, NULL, &info, &error_abort);
2748     visit_complete(v, &obj);
2749     str = qobject_to_json_pretty(obj, true);
2750     assert(str != NULL);
2751     printf("%s\n", str->str);
2752     qobject_unref(obj);
2753     visit_free(v);
2754     g_string_free(str, true);
2755 }
2756 
2757 static void dump_human_image_info_list(ImageInfoList *list)
2758 {
2759     ImageInfoList *elem;
2760     bool delim = false;
2761 
2762     for (elem = list; elem; elem = elem->next) {
2763         if (delim) {
2764             printf("\n");
2765         }
2766         delim = true;
2767 
2768         bdrv_image_info_dump(elem->value);
2769     }
2770 }
2771 
2772 static gboolean str_equal_func(gconstpointer a, gconstpointer b)
2773 {
2774     return strcmp(a, b) == 0;
2775 }
2776 
2777 /**
2778  * Open an image file chain and return an ImageInfoList
2779  *
2780  * @filename: topmost image filename
2781  * @fmt: topmost image format (may be NULL to autodetect)
2782  * @chain: true  - enumerate entire backing file chain
2783  *         false - only topmost image file
2784  *
2785  * Returns a list of ImageInfo objects or NULL if there was an error opening an
2786  * image file.  If there was an error a message will have been printed to
2787  * stderr.
2788  */
2789 static ImageInfoList *collect_image_info_list(bool image_opts,
2790                                               const char *filename,
2791                                               const char *fmt,
2792                                               bool chain, bool force_share)
2793 {
2794     ImageInfoList *head = NULL;
2795     ImageInfoList **tail = &head;
2796     GHashTable *filenames;
2797     Error *err = NULL;
2798 
2799     filenames = g_hash_table_new_full(g_str_hash, str_equal_func, NULL, NULL);
2800 
2801     while (filename) {
2802         BlockBackend *blk;
2803         BlockDriverState *bs;
2804         ImageInfo *info;
2805 
2806         if (g_hash_table_lookup_extended(filenames, filename, NULL, NULL)) {
2807             error_report("Backing file '%s' creates an infinite loop.",
2808                          filename);
2809             goto err;
2810         }
2811         g_hash_table_insert(filenames, (gpointer)filename, NULL);
2812 
2813         blk = img_open(image_opts, filename, fmt,
2814                        BDRV_O_NO_BACKING | BDRV_O_NO_IO, false, false,
2815                        force_share);
2816         if (!blk) {
2817             goto err;
2818         }
2819         bs = blk_bs(blk);
2820 
2821         bdrv_query_image_info(bs, &info, &err);
2822         if (err) {
2823             error_report_err(err);
2824             blk_unref(blk);
2825             goto err;
2826         }
2827 
2828         QAPI_LIST_APPEND(tail, info);
2829 
2830         blk_unref(blk);
2831 
2832         /* Clear parameters that only apply to the topmost image */
2833         filename = fmt = NULL;
2834         image_opts = false;
2835 
2836         if (chain) {
2837             if (info->has_full_backing_filename) {
2838                 filename = info->full_backing_filename;
2839             } else if (info->has_backing_filename) {
2840                 error_report("Could not determine absolute backing filename,"
2841                              " but backing filename '%s' present",
2842                              info->backing_filename);
2843                 goto err;
2844             }
2845             if (info->has_backing_filename_format) {
2846                 fmt = info->backing_filename_format;
2847             }
2848         }
2849     }
2850     g_hash_table_destroy(filenames);
2851     return head;
2852 
2853 err:
2854     qapi_free_ImageInfoList(head);
2855     g_hash_table_destroy(filenames);
2856     return NULL;
2857 }
2858 
2859 static int img_info(int argc, char **argv)
2860 {
2861     int c;
2862     OutputFormat output_format = OFORMAT_HUMAN;
2863     bool chain = false;
2864     const char *filename, *fmt, *output;
2865     ImageInfoList *list;
2866     bool image_opts = false;
2867     bool force_share = false;
2868 
2869     fmt = NULL;
2870     output = NULL;
2871     for(;;) {
2872         int option_index = 0;
2873         static const struct option long_options[] = {
2874             {"help", no_argument, 0, 'h'},
2875             {"format", required_argument, 0, 'f'},
2876             {"output", required_argument, 0, OPTION_OUTPUT},
2877             {"backing-chain", no_argument, 0, OPTION_BACKING_CHAIN},
2878             {"object", required_argument, 0, OPTION_OBJECT},
2879             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2880             {"force-share", no_argument, 0, 'U'},
2881             {0, 0, 0, 0}
2882         };
2883         c = getopt_long(argc, argv, ":f:hU",
2884                         long_options, &option_index);
2885         if (c == -1) {
2886             break;
2887         }
2888         switch(c) {
2889         case ':':
2890             missing_argument(argv[optind - 1]);
2891             break;
2892         case '?':
2893             unrecognized_option(argv[optind - 1]);
2894             break;
2895         case 'h':
2896             help();
2897             break;
2898         case 'f':
2899             fmt = optarg;
2900             break;
2901         case 'U':
2902             force_share = true;
2903             break;
2904         case OPTION_OUTPUT:
2905             output = optarg;
2906             break;
2907         case OPTION_BACKING_CHAIN:
2908             chain = true;
2909             break;
2910         case OPTION_OBJECT:
2911             user_creatable_process_cmdline(optarg);
2912             break;
2913         case OPTION_IMAGE_OPTS:
2914             image_opts = true;
2915             break;
2916         }
2917     }
2918     if (optind != argc - 1) {
2919         error_exit("Expecting one image file name");
2920     }
2921     filename = argv[optind++];
2922 
2923     if (output && !strcmp(output, "json")) {
2924         output_format = OFORMAT_JSON;
2925     } else if (output && !strcmp(output, "human")) {
2926         output_format = OFORMAT_HUMAN;
2927     } else if (output) {
2928         error_report("--output must be used with human or json as argument.");
2929         return 1;
2930     }
2931 
2932     list = collect_image_info_list(image_opts, filename, fmt, chain,
2933                                    force_share);
2934     if (!list) {
2935         return 1;
2936     }
2937 
2938     switch (output_format) {
2939     case OFORMAT_HUMAN:
2940         dump_human_image_info_list(list);
2941         break;
2942     case OFORMAT_JSON:
2943         if (chain) {
2944             dump_json_image_info_list(list);
2945         } else {
2946             dump_json_image_info(list->value);
2947         }
2948         break;
2949     }
2950 
2951     qapi_free_ImageInfoList(list);
2952     return 0;
2953 }
2954 
2955 static int dump_map_entry(OutputFormat output_format, MapEntry *e,
2956                           MapEntry *next)
2957 {
2958     switch (output_format) {
2959     case OFORMAT_HUMAN:
2960         if (e->data && !e->has_offset) {
2961             error_report("File contains external, encrypted or compressed clusters.");
2962             return -1;
2963         }
2964         if (e->data && !e->zero) {
2965             printf("%#-16"PRIx64"%#-16"PRIx64"%#-16"PRIx64"%s\n",
2966                    e->start, e->length,
2967                    e->has_offset ? e->offset : 0,
2968                    e->has_filename ? e->filename : "");
2969         }
2970         /* This format ignores the distinction between 0, ZERO and ZERO|DATA.
2971          * Modify the flags here to allow more coalescing.
2972          */
2973         if (next && (!next->data || next->zero)) {
2974             next->data = false;
2975             next->zero = true;
2976         }
2977         break;
2978     case OFORMAT_JSON:
2979         printf("{ \"start\": %"PRId64", \"length\": %"PRId64","
2980                " \"depth\": %"PRId64", \"zero\": %s, \"data\": %s",
2981                e->start, e->length, e->depth,
2982                e->zero ? "true" : "false",
2983                e->data ? "true" : "false");
2984         if (e->has_offset) {
2985             printf(", \"offset\": %"PRId64"", e->offset);
2986         }
2987         putchar('}');
2988 
2989         if (next) {
2990             puts(",");
2991         }
2992         break;
2993     }
2994     return 0;
2995 }
2996 
2997 static int get_block_status(BlockDriverState *bs, int64_t offset,
2998                             int64_t bytes, MapEntry *e)
2999 {
3000     int ret;
3001     int depth;
3002     BlockDriverState *file;
3003     bool has_offset;
3004     int64_t map;
3005     char *filename = NULL;
3006 
3007     /* As an optimization, we could cache the current range of unallocated
3008      * clusters in each file of the chain, and avoid querying the same
3009      * range repeatedly.
3010      */
3011 
3012     depth = 0;
3013     for (;;) {
3014         bs = bdrv_skip_filters(bs);
3015         ret = bdrv_block_status(bs, offset, bytes, &bytes, &map, &file);
3016         if (ret < 0) {
3017             return ret;
3018         }
3019         assert(bytes);
3020         if (ret & (BDRV_BLOCK_ZERO|BDRV_BLOCK_DATA)) {
3021             break;
3022         }
3023         bs = bdrv_cow_bs(bs);
3024         if (bs == NULL) {
3025             ret = 0;
3026             break;
3027         }
3028 
3029         depth++;
3030     }
3031 
3032     has_offset = !!(ret & BDRV_BLOCK_OFFSET_VALID);
3033 
3034     if (file && has_offset) {
3035         bdrv_refresh_filename(file);
3036         filename = file->filename;
3037     }
3038 
3039     *e = (MapEntry) {
3040         .start = offset,
3041         .length = bytes,
3042         .data = !!(ret & BDRV_BLOCK_DATA),
3043         .zero = !!(ret & BDRV_BLOCK_ZERO),
3044         .offset = map,
3045         .has_offset = has_offset,
3046         .depth = depth,
3047         .has_filename = filename,
3048         .filename = filename,
3049     };
3050 
3051     return 0;
3052 }
3053 
3054 static inline bool entry_mergeable(const MapEntry *curr, const MapEntry *next)
3055 {
3056     if (curr->length == 0) {
3057         return false;
3058     }
3059     if (curr->zero != next->zero ||
3060         curr->data != next->data ||
3061         curr->depth != next->depth ||
3062         curr->has_filename != next->has_filename ||
3063         curr->has_offset != next->has_offset) {
3064         return false;
3065     }
3066     if (curr->has_filename && strcmp(curr->filename, next->filename)) {
3067         return false;
3068     }
3069     if (curr->has_offset && curr->offset + curr->length != next->offset) {
3070         return false;
3071     }
3072     return true;
3073 }
3074 
3075 static int img_map(int argc, char **argv)
3076 {
3077     int c;
3078     OutputFormat output_format = OFORMAT_HUMAN;
3079     BlockBackend *blk;
3080     BlockDriverState *bs;
3081     const char *filename, *fmt, *output;
3082     int64_t length;
3083     MapEntry curr = { .length = 0 }, next;
3084     int ret = 0;
3085     bool image_opts = false;
3086     bool force_share = false;
3087     int64_t start_offset = 0;
3088     int64_t max_length = -1;
3089 
3090     fmt = NULL;
3091     output = NULL;
3092     for (;;) {
3093         int option_index = 0;
3094         static const struct option long_options[] = {
3095             {"help", no_argument, 0, 'h'},
3096             {"format", required_argument, 0, 'f'},
3097             {"output", required_argument, 0, OPTION_OUTPUT},
3098             {"object", required_argument, 0, OPTION_OBJECT},
3099             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3100             {"force-share", no_argument, 0, 'U'},
3101             {"start-offset", required_argument, 0, 's'},
3102             {"max-length", required_argument, 0, 'l'},
3103             {0, 0, 0, 0}
3104         };
3105         c = getopt_long(argc, argv, ":f:s:l:hU",
3106                         long_options, &option_index);
3107         if (c == -1) {
3108             break;
3109         }
3110         switch (c) {
3111         case ':':
3112             missing_argument(argv[optind - 1]);
3113             break;
3114         case '?':
3115             unrecognized_option(argv[optind - 1]);
3116             break;
3117         case 'h':
3118             help();
3119             break;
3120         case 'f':
3121             fmt = optarg;
3122             break;
3123         case 'U':
3124             force_share = true;
3125             break;
3126         case OPTION_OUTPUT:
3127             output = optarg;
3128             break;
3129         case 's':
3130             start_offset = cvtnum("start offset", optarg);
3131             if (start_offset < 0) {
3132                 return 1;
3133             }
3134             break;
3135         case 'l':
3136             max_length = cvtnum("max length", optarg);
3137             if (max_length < 0) {
3138                 return 1;
3139             }
3140             break;
3141         case OPTION_OBJECT:
3142             user_creatable_process_cmdline(optarg);
3143             break;
3144         case OPTION_IMAGE_OPTS:
3145             image_opts = true;
3146             break;
3147         }
3148     }
3149     if (optind != argc - 1) {
3150         error_exit("Expecting one image file name");
3151     }
3152     filename = argv[optind];
3153 
3154     if (output && !strcmp(output, "json")) {
3155         output_format = OFORMAT_JSON;
3156     } else if (output && !strcmp(output, "human")) {
3157         output_format = OFORMAT_HUMAN;
3158     } else if (output) {
3159         error_report("--output must be used with human or json as argument.");
3160         return 1;
3161     }
3162 
3163     blk = img_open(image_opts, filename, fmt, 0, false, false, force_share);
3164     if (!blk) {
3165         return 1;
3166     }
3167     bs = blk_bs(blk);
3168 
3169     if (output_format == OFORMAT_HUMAN) {
3170         printf("%-16s%-16s%-16s%s\n", "Offset", "Length", "Mapped to", "File");
3171     } else if (output_format == OFORMAT_JSON) {
3172         putchar('[');
3173     }
3174 
3175     length = blk_getlength(blk);
3176     if (length < 0) {
3177         error_report("Failed to get size for '%s'", filename);
3178         return 1;
3179     }
3180     if (max_length != -1) {
3181         length = MIN(start_offset + max_length, length);
3182     }
3183 
3184     curr.start = start_offset;
3185     while (curr.start + curr.length < length) {
3186         int64_t offset = curr.start + curr.length;
3187         int64_t n = length - offset;
3188 
3189         ret = get_block_status(bs, offset, n, &next);
3190         if (ret < 0) {
3191             error_report("Could not read file metadata: %s", strerror(-ret));
3192             goto out;
3193         }
3194 
3195         if (entry_mergeable(&curr, &next)) {
3196             curr.length += next.length;
3197             continue;
3198         }
3199 
3200         if (curr.length > 0) {
3201             ret = dump_map_entry(output_format, &curr, &next);
3202             if (ret < 0) {
3203                 goto out;
3204             }
3205         }
3206         curr = next;
3207     }
3208 
3209     ret = dump_map_entry(output_format, &curr, NULL);
3210     if (output_format == OFORMAT_JSON) {
3211         puts("]");
3212     }
3213 
3214 out:
3215     blk_unref(blk);
3216     return ret < 0;
3217 }
3218 
3219 #define SNAPSHOT_LIST   1
3220 #define SNAPSHOT_CREATE 2
3221 #define SNAPSHOT_APPLY  3
3222 #define SNAPSHOT_DELETE 4
3223 
3224 static int img_snapshot(int argc, char **argv)
3225 {
3226     BlockBackend *blk;
3227     BlockDriverState *bs;
3228     QEMUSnapshotInfo sn;
3229     char *filename, *snapshot_name = NULL;
3230     int c, ret = 0, bdrv_oflags;
3231     int action = 0;
3232     qemu_timeval tv;
3233     bool quiet = false;
3234     Error *err = NULL;
3235     bool image_opts = false;
3236     bool force_share = false;
3237 
3238     bdrv_oflags = BDRV_O_RDWR;
3239     /* Parse commandline parameters */
3240     for(;;) {
3241         static const struct option long_options[] = {
3242             {"help", no_argument, 0, 'h'},
3243             {"object", required_argument, 0, OPTION_OBJECT},
3244             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3245             {"force-share", no_argument, 0, 'U'},
3246             {0, 0, 0, 0}
3247         };
3248         c = getopt_long(argc, argv, ":la:c:d:hqU",
3249                         long_options, NULL);
3250         if (c == -1) {
3251             break;
3252         }
3253         switch(c) {
3254         case ':':
3255             missing_argument(argv[optind - 1]);
3256             break;
3257         case '?':
3258             unrecognized_option(argv[optind - 1]);
3259             break;
3260         case 'h':
3261             help();
3262             return 0;
3263         case 'l':
3264             if (action) {
3265                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3266                 return 0;
3267             }
3268             action = SNAPSHOT_LIST;
3269             bdrv_oflags &= ~BDRV_O_RDWR; /* no need for RW */
3270             break;
3271         case 'a':
3272             if (action) {
3273                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3274                 return 0;
3275             }
3276             action = SNAPSHOT_APPLY;
3277             snapshot_name = optarg;
3278             break;
3279         case 'c':
3280             if (action) {
3281                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3282                 return 0;
3283             }
3284             action = SNAPSHOT_CREATE;
3285             snapshot_name = optarg;
3286             break;
3287         case 'd':
3288             if (action) {
3289                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3290                 return 0;
3291             }
3292             action = SNAPSHOT_DELETE;
3293             snapshot_name = optarg;
3294             break;
3295         case 'q':
3296             quiet = true;
3297             break;
3298         case 'U':
3299             force_share = true;
3300             break;
3301         case OPTION_OBJECT:
3302             user_creatable_process_cmdline(optarg);
3303             break;
3304         case OPTION_IMAGE_OPTS:
3305             image_opts = true;
3306             break;
3307         }
3308     }
3309 
3310     if (optind != argc - 1) {
3311         error_exit("Expecting one image file name");
3312     }
3313     filename = argv[optind++];
3314 
3315     /* Open the image */
3316     blk = img_open(image_opts, filename, NULL, bdrv_oflags, false, quiet,
3317                    force_share);
3318     if (!blk) {
3319         return 1;
3320     }
3321     bs = blk_bs(blk);
3322 
3323     /* Perform the requested action */
3324     switch(action) {
3325     case SNAPSHOT_LIST:
3326         dump_snapshots(bs);
3327         break;
3328 
3329     case SNAPSHOT_CREATE:
3330         memset(&sn, 0, sizeof(sn));
3331         pstrcpy(sn.name, sizeof(sn.name), snapshot_name);
3332 
3333         qemu_gettimeofday(&tv);
3334         sn.date_sec = tv.tv_sec;
3335         sn.date_nsec = tv.tv_usec * 1000;
3336 
3337         ret = bdrv_snapshot_create(bs, &sn);
3338         if (ret) {
3339             error_report("Could not create snapshot '%s': %d (%s)",
3340                 snapshot_name, ret, strerror(-ret));
3341         }
3342         break;
3343 
3344     case SNAPSHOT_APPLY:
3345         ret = bdrv_snapshot_goto(bs, snapshot_name, &err);
3346         if (ret) {
3347             error_reportf_err(err, "Could not apply snapshot '%s': ",
3348                               snapshot_name);
3349         }
3350         break;
3351 
3352     case SNAPSHOT_DELETE:
3353         ret = bdrv_snapshot_find(bs, &sn, snapshot_name);
3354         if (ret < 0) {
3355             error_report("Could not delete snapshot '%s': snapshot not "
3356                          "found", snapshot_name);
3357             ret = 1;
3358         } else {
3359             ret = bdrv_snapshot_delete(bs, sn.id_str, sn.name, &err);
3360             if (ret < 0) {
3361                 error_reportf_err(err, "Could not delete snapshot '%s': ",
3362                                   snapshot_name);
3363                 ret = 1;
3364             }
3365         }
3366         break;
3367     }
3368 
3369     /* Cleanup */
3370     blk_unref(blk);
3371     if (ret) {
3372         return 1;
3373     }
3374     return 0;
3375 }
3376 
3377 static int img_rebase(int argc, char **argv)
3378 {
3379     BlockBackend *blk = NULL, *blk_old_backing = NULL, *blk_new_backing = NULL;
3380     uint8_t *buf_old = NULL;
3381     uint8_t *buf_new = NULL;
3382     BlockDriverState *bs = NULL, *prefix_chain_bs = NULL;
3383     BlockDriverState *unfiltered_bs;
3384     char *filename;
3385     const char *fmt, *cache, *src_cache, *out_basefmt, *out_baseimg;
3386     int c, flags, src_flags, ret;
3387     bool writethrough, src_writethrough;
3388     int unsafe = 0;
3389     bool force_share = false;
3390     int progress = 0;
3391     bool quiet = false;
3392     Error *local_err = NULL;
3393     bool image_opts = false;
3394 
3395     /* Parse commandline parameters */
3396     fmt = NULL;
3397     cache = BDRV_DEFAULT_CACHE;
3398     src_cache = BDRV_DEFAULT_CACHE;
3399     out_baseimg = NULL;
3400     out_basefmt = NULL;
3401     for(;;) {
3402         static const struct option long_options[] = {
3403             {"help", no_argument, 0, 'h'},
3404             {"object", required_argument, 0, OPTION_OBJECT},
3405             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3406             {"force-share", no_argument, 0, 'U'},
3407             {0, 0, 0, 0}
3408         };
3409         c = getopt_long(argc, argv, ":hf:F:b:upt:T:qU",
3410                         long_options, NULL);
3411         if (c == -1) {
3412             break;
3413         }
3414         switch(c) {
3415         case ':':
3416             missing_argument(argv[optind - 1]);
3417             break;
3418         case '?':
3419             unrecognized_option(argv[optind - 1]);
3420             break;
3421         case 'h':
3422             help();
3423             return 0;
3424         case 'f':
3425             fmt = optarg;
3426             break;
3427         case 'F':
3428             out_basefmt = optarg;
3429             break;
3430         case 'b':
3431             out_baseimg = optarg;
3432             break;
3433         case 'u':
3434             unsafe = 1;
3435             break;
3436         case 'p':
3437             progress = 1;
3438             break;
3439         case 't':
3440             cache = optarg;
3441             break;
3442         case 'T':
3443             src_cache = optarg;
3444             break;
3445         case 'q':
3446             quiet = true;
3447             break;
3448         case OPTION_OBJECT:
3449             user_creatable_process_cmdline(optarg);
3450             break;
3451         case OPTION_IMAGE_OPTS:
3452             image_opts = true;
3453             break;
3454         case 'U':
3455             force_share = true;
3456             break;
3457         }
3458     }
3459 
3460     if (quiet) {
3461         progress = 0;
3462     }
3463 
3464     if (optind != argc - 1) {
3465         error_exit("Expecting one image file name");
3466     }
3467     if (!unsafe && !out_baseimg) {
3468         error_exit("Must specify backing file (-b) or use unsafe mode (-u)");
3469     }
3470     filename = argv[optind++];
3471 
3472     qemu_progress_init(progress, 2.0);
3473     qemu_progress_print(0, 100);
3474 
3475     flags = BDRV_O_RDWR | (unsafe ? BDRV_O_NO_BACKING : 0);
3476     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
3477     if (ret < 0) {
3478         error_report("Invalid cache option: %s", cache);
3479         goto out;
3480     }
3481 
3482     src_flags = 0;
3483     ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
3484     if (ret < 0) {
3485         error_report("Invalid source cache option: %s", src_cache);
3486         goto out;
3487     }
3488 
3489     /* The source files are opened read-only, don't care about WCE */
3490     assert((src_flags & BDRV_O_RDWR) == 0);
3491     (void) src_writethrough;
3492 
3493     /*
3494      * Open the images.
3495      *
3496      * Ignore the old backing file for unsafe rebase in case we want to correct
3497      * the reference to a renamed or moved backing file.
3498      */
3499     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
3500                    false);
3501     if (!blk) {
3502         ret = -1;
3503         goto out;
3504     }
3505     bs = blk_bs(blk);
3506 
3507     unfiltered_bs = bdrv_skip_filters(bs);
3508 
3509     if (out_basefmt != NULL) {
3510         if (bdrv_find_format(out_basefmt) == NULL) {
3511             error_report("Invalid format name: '%s'", out_basefmt);
3512             ret = -1;
3513             goto out;
3514         }
3515     }
3516 
3517     /* For safe rebasing we need to compare old and new backing file */
3518     if (!unsafe) {
3519         QDict *options = NULL;
3520         BlockDriverState *base_bs = bdrv_cow_bs(unfiltered_bs);
3521 
3522         if (base_bs) {
3523             blk_old_backing = blk_new(qemu_get_aio_context(),
3524                                       BLK_PERM_CONSISTENT_READ,
3525                                       BLK_PERM_ALL);
3526             ret = blk_insert_bs(blk_old_backing, base_bs,
3527                                 &local_err);
3528             if (ret < 0) {
3529                 error_reportf_err(local_err,
3530                                   "Could not reuse old backing file '%s': ",
3531                                   base_bs->filename);
3532                 goto out;
3533             }
3534         } else {
3535             blk_old_backing = NULL;
3536         }
3537 
3538         if (out_baseimg[0]) {
3539             const char *overlay_filename;
3540             char *out_real_path;
3541 
3542             options = qdict_new();
3543             if (out_basefmt) {
3544                 qdict_put_str(options, "driver", out_basefmt);
3545             }
3546             if (force_share) {
3547                 qdict_put_bool(options, BDRV_OPT_FORCE_SHARE, true);
3548             }
3549 
3550             bdrv_refresh_filename(bs);
3551             overlay_filename = bs->exact_filename[0] ? bs->exact_filename
3552                                                      : bs->filename;
3553             out_real_path =
3554                 bdrv_get_full_backing_filename_from_filename(overlay_filename,
3555                                                              out_baseimg,
3556                                                              &local_err);
3557             if (local_err) {
3558                 qobject_unref(options);
3559                 error_reportf_err(local_err,
3560                                   "Could not resolve backing filename: ");
3561                 ret = -1;
3562                 goto out;
3563             }
3564 
3565             /*
3566              * Find out whether we rebase an image on top of a previous image
3567              * in its chain.
3568              */
3569             prefix_chain_bs = bdrv_find_backing_image(bs, out_real_path);
3570             if (prefix_chain_bs) {
3571                 qobject_unref(options);
3572                 g_free(out_real_path);
3573 
3574                 blk_new_backing = blk_new(qemu_get_aio_context(),
3575                                           BLK_PERM_CONSISTENT_READ,
3576                                           BLK_PERM_ALL);
3577                 ret = blk_insert_bs(blk_new_backing, prefix_chain_bs,
3578                                     &local_err);
3579                 if (ret < 0) {
3580                     error_reportf_err(local_err,
3581                                       "Could not reuse backing file '%s': ",
3582                                       out_baseimg);
3583                     goto out;
3584                 }
3585             } else {
3586                 blk_new_backing = blk_new_open(out_real_path, NULL,
3587                                                options, src_flags, &local_err);
3588                 g_free(out_real_path);
3589                 if (!blk_new_backing) {
3590                     error_reportf_err(local_err,
3591                                       "Could not open new backing file '%s': ",
3592                                       out_baseimg);
3593                     ret = -1;
3594                     goto out;
3595                 }
3596             }
3597         }
3598     }
3599 
3600     /*
3601      * Check each unallocated cluster in the COW file. If it is unallocated,
3602      * accesses go to the backing file. We must therefore compare this cluster
3603      * in the old and new backing file, and if they differ we need to copy it
3604      * from the old backing file into the COW file.
3605      *
3606      * If qemu-img crashes during this step, no harm is done. The content of
3607      * the image is the same as the original one at any time.
3608      */
3609     if (!unsafe) {
3610         int64_t size;
3611         int64_t old_backing_size = 0;
3612         int64_t new_backing_size = 0;
3613         uint64_t offset;
3614         int64_t n;
3615         float local_progress = 0;
3616 
3617         buf_old = blk_blockalign(blk, IO_BUF_SIZE);
3618         buf_new = blk_blockalign(blk, IO_BUF_SIZE);
3619 
3620         size = blk_getlength(blk);
3621         if (size < 0) {
3622             error_report("Could not get size of '%s': %s",
3623                          filename, strerror(-size));
3624             ret = -1;
3625             goto out;
3626         }
3627         if (blk_old_backing) {
3628             old_backing_size = blk_getlength(blk_old_backing);
3629             if (old_backing_size < 0) {
3630                 char backing_name[PATH_MAX];
3631 
3632                 bdrv_get_backing_filename(bs, backing_name,
3633                                           sizeof(backing_name));
3634                 error_report("Could not get size of '%s': %s",
3635                              backing_name, strerror(-old_backing_size));
3636                 ret = -1;
3637                 goto out;
3638             }
3639         }
3640         if (blk_new_backing) {
3641             new_backing_size = blk_getlength(blk_new_backing);
3642             if (new_backing_size < 0) {
3643                 error_report("Could not get size of '%s': %s",
3644                              out_baseimg, strerror(-new_backing_size));
3645                 ret = -1;
3646                 goto out;
3647             }
3648         }
3649 
3650         if (size != 0) {
3651             local_progress = (float)100 / (size / MIN(size, IO_BUF_SIZE));
3652         }
3653 
3654         for (offset = 0; offset < size; offset += n) {
3655             bool buf_old_is_zero = false;
3656 
3657             /* How many bytes can we handle with the next read? */
3658             n = MIN(IO_BUF_SIZE, size - offset);
3659 
3660             /* If the cluster is allocated, we don't need to take action */
3661             ret = bdrv_is_allocated(unfiltered_bs, offset, n, &n);
3662             if (ret < 0) {
3663                 error_report("error while reading image metadata: %s",
3664                              strerror(-ret));
3665                 goto out;
3666             }
3667             if (ret) {
3668                 continue;
3669             }
3670 
3671             if (prefix_chain_bs) {
3672                 /*
3673                  * If cluster wasn't changed since prefix_chain, we don't need
3674                  * to take action
3675                  */
3676                 ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
3677                                               prefix_chain_bs, false,
3678                                               offset, n, &n);
3679                 if (ret < 0) {
3680                     error_report("error while reading image metadata: %s",
3681                                  strerror(-ret));
3682                     goto out;
3683                 }
3684                 if (!ret) {
3685                     continue;
3686                 }
3687             }
3688 
3689             /*
3690              * Read old and new backing file and take into consideration that
3691              * backing files may be smaller than the COW image.
3692              */
3693             if (offset >= old_backing_size) {
3694                 memset(buf_old, 0, n);
3695                 buf_old_is_zero = true;
3696             } else {
3697                 if (offset + n > old_backing_size) {
3698                     n = old_backing_size - offset;
3699                 }
3700 
3701                 ret = blk_pread(blk_old_backing, offset, buf_old, n);
3702                 if (ret < 0) {
3703                     error_report("error while reading from old backing file");
3704                     goto out;
3705                 }
3706             }
3707 
3708             if (offset >= new_backing_size || !blk_new_backing) {
3709                 memset(buf_new, 0, n);
3710             } else {
3711                 if (offset + n > new_backing_size) {
3712                     n = new_backing_size - offset;
3713                 }
3714 
3715                 ret = blk_pread(blk_new_backing, offset, buf_new, n);
3716                 if (ret < 0) {
3717                     error_report("error while reading from new backing file");
3718                     goto out;
3719                 }
3720             }
3721 
3722             /* If they differ, we need to write to the COW file */
3723             uint64_t written = 0;
3724 
3725             while (written < n) {
3726                 int64_t pnum;
3727 
3728                 if (compare_buffers(buf_old + written, buf_new + written,
3729                                     n - written, &pnum))
3730                 {
3731                     if (buf_old_is_zero) {
3732                         ret = blk_pwrite_zeroes(blk, offset + written, pnum, 0);
3733                     } else {
3734                         ret = blk_pwrite(blk, offset + written,
3735                                          buf_old + written, pnum, 0);
3736                     }
3737                     if (ret < 0) {
3738                         error_report("Error while writing to COW image: %s",
3739                             strerror(-ret));
3740                         goto out;
3741                     }
3742                 }
3743 
3744                 written += pnum;
3745             }
3746             qemu_progress_print(local_progress, 100);
3747         }
3748     }
3749 
3750     /*
3751      * Change the backing file. All clusters that are different from the old
3752      * backing file are overwritten in the COW file now, so the visible content
3753      * doesn't change when we switch the backing file.
3754      */
3755     if (out_baseimg && *out_baseimg) {
3756         ret = bdrv_change_backing_file(unfiltered_bs, out_baseimg, out_basefmt,
3757                                        true);
3758     } else {
3759         ret = bdrv_change_backing_file(unfiltered_bs, NULL, NULL, false);
3760     }
3761 
3762     if (ret == -ENOSPC) {
3763         error_report("Could not change the backing file to '%s': No "
3764                      "space left in the file header", out_baseimg);
3765     } else if (ret < 0) {
3766         error_report("Could not change the backing file to '%s': %s",
3767             out_baseimg, strerror(-ret));
3768     }
3769 
3770     qemu_progress_print(100, 0);
3771     /*
3772      * TODO At this point it is possible to check if any clusters that are
3773      * allocated in the COW file are the same in the backing file. If so, they
3774      * could be dropped from the COW file. Don't do this before switching the
3775      * backing file, in case of a crash this would lead to corruption.
3776      */
3777 out:
3778     qemu_progress_end();
3779     /* Cleanup */
3780     if (!unsafe) {
3781         blk_unref(blk_old_backing);
3782         blk_unref(blk_new_backing);
3783     }
3784     qemu_vfree(buf_old);
3785     qemu_vfree(buf_new);
3786 
3787     blk_unref(blk);
3788     if (ret) {
3789         return 1;
3790     }
3791     return 0;
3792 }
3793 
3794 static int img_resize(int argc, char **argv)
3795 {
3796     Error *err = NULL;
3797     int c, ret, relative;
3798     const char *filename, *fmt, *size;
3799     int64_t n, total_size, current_size;
3800     bool quiet = false;
3801     BlockBackend *blk = NULL;
3802     PreallocMode prealloc = PREALLOC_MODE_OFF;
3803     QemuOpts *param;
3804 
3805     static QemuOptsList resize_options = {
3806         .name = "resize_options",
3807         .head = QTAILQ_HEAD_INITIALIZER(resize_options.head),
3808         .desc = {
3809             {
3810                 .name = BLOCK_OPT_SIZE,
3811                 .type = QEMU_OPT_SIZE,
3812                 .help = "Virtual disk size"
3813             }, {
3814                 /* end of list */
3815             }
3816         },
3817     };
3818     bool image_opts = false;
3819     bool shrink = false;
3820 
3821     /* Remove size from argv manually so that negative numbers are not treated
3822      * as options by getopt. */
3823     if (argc < 3) {
3824         error_exit("Not enough arguments");
3825         return 1;
3826     }
3827 
3828     size = argv[--argc];
3829 
3830     /* Parse getopt arguments */
3831     fmt = NULL;
3832     for(;;) {
3833         static const struct option long_options[] = {
3834             {"help", no_argument, 0, 'h'},
3835             {"object", required_argument, 0, OPTION_OBJECT},
3836             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3837             {"preallocation", required_argument, 0, OPTION_PREALLOCATION},
3838             {"shrink", no_argument, 0, OPTION_SHRINK},
3839             {0, 0, 0, 0}
3840         };
3841         c = getopt_long(argc, argv, ":f:hq",
3842                         long_options, NULL);
3843         if (c == -1) {
3844             break;
3845         }
3846         switch(c) {
3847         case ':':
3848             missing_argument(argv[optind - 1]);
3849             break;
3850         case '?':
3851             unrecognized_option(argv[optind - 1]);
3852             break;
3853         case 'h':
3854             help();
3855             break;
3856         case 'f':
3857             fmt = optarg;
3858             break;
3859         case 'q':
3860             quiet = true;
3861             break;
3862         case OPTION_OBJECT:
3863             user_creatable_process_cmdline(optarg);
3864             break;
3865         case OPTION_IMAGE_OPTS:
3866             image_opts = true;
3867             break;
3868         case OPTION_PREALLOCATION:
3869             prealloc = qapi_enum_parse(&PreallocMode_lookup, optarg,
3870                                        PREALLOC_MODE__MAX, NULL);
3871             if (prealloc == PREALLOC_MODE__MAX) {
3872                 error_report("Invalid preallocation mode '%s'", optarg);
3873                 return 1;
3874             }
3875             break;
3876         case OPTION_SHRINK:
3877             shrink = true;
3878             break;
3879         }
3880     }
3881     if (optind != argc - 1) {
3882         error_exit("Expecting image file name and size");
3883     }
3884     filename = argv[optind++];
3885 
3886     /* Choose grow, shrink, or absolute resize mode */
3887     switch (size[0]) {
3888     case '+':
3889         relative = 1;
3890         size++;
3891         break;
3892     case '-':
3893         relative = -1;
3894         size++;
3895         break;
3896     default:
3897         relative = 0;
3898         break;
3899     }
3900 
3901     /* Parse size */
3902     param = qemu_opts_create(&resize_options, NULL, 0, &error_abort);
3903     if (!qemu_opt_set(param, BLOCK_OPT_SIZE, size, &err)) {
3904         error_report_err(err);
3905         ret = -1;
3906         qemu_opts_del(param);
3907         goto out;
3908     }
3909     n = qemu_opt_get_size(param, BLOCK_OPT_SIZE, 0);
3910     qemu_opts_del(param);
3911 
3912     blk = img_open(image_opts, filename, fmt,
3913                    BDRV_O_RDWR | BDRV_O_RESIZE, false, quiet,
3914                    false);
3915     if (!blk) {
3916         ret = -1;
3917         goto out;
3918     }
3919 
3920     current_size = blk_getlength(blk);
3921     if (current_size < 0) {
3922         error_report("Failed to inquire current image length: %s",
3923                      strerror(-current_size));
3924         ret = -1;
3925         goto out;
3926     }
3927 
3928     if (relative) {
3929         total_size = current_size + n * relative;
3930     } else {
3931         total_size = n;
3932     }
3933     if (total_size <= 0) {
3934         error_report("New image size must be positive");
3935         ret = -1;
3936         goto out;
3937     }
3938 
3939     if (total_size <= current_size && prealloc != PREALLOC_MODE_OFF) {
3940         error_report("Preallocation can only be used for growing images");
3941         ret = -1;
3942         goto out;
3943     }
3944 
3945     if (total_size < current_size && !shrink) {
3946         error_report("Use the --shrink option to perform a shrink operation.");
3947         warn_report("Shrinking an image will delete all data beyond the "
3948                     "shrunken image's end. Before performing such an "
3949                     "operation, make sure there is no important data there.");
3950         ret = -1;
3951         goto out;
3952     }
3953 
3954     /*
3955      * The user expects the image to have the desired size after
3956      * resizing, so pass @exact=true.  It is of no use to report
3957      * success when the image has not actually been resized.
3958      */
3959     ret = blk_truncate(blk, total_size, true, prealloc, 0, &err);
3960     if (!ret) {
3961         qprintf(quiet, "Image resized.\n");
3962     } else {
3963         error_report_err(err);
3964     }
3965 out:
3966     blk_unref(blk);
3967     if (ret) {
3968         return 1;
3969     }
3970     return 0;
3971 }
3972 
3973 static void amend_status_cb(BlockDriverState *bs,
3974                             int64_t offset, int64_t total_work_size,
3975                             void *opaque)
3976 {
3977     qemu_progress_print(100.f * offset / total_work_size, 0);
3978 }
3979 
3980 static int print_amend_option_help(const char *format)
3981 {
3982     BlockDriver *drv;
3983 
3984     /* Find driver and parse its options */
3985     drv = bdrv_find_format(format);
3986     if (!drv) {
3987         error_report("Unknown file format '%s'", format);
3988         return 1;
3989     }
3990 
3991     if (!drv->bdrv_amend_options) {
3992         error_report("Format driver '%s' does not support option amendment",
3993                      format);
3994         return 1;
3995     }
3996 
3997     /* Every driver supporting amendment must have amend_opts */
3998     assert(drv->amend_opts);
3999 
4000     printf("Amend options for '%s':\n", format);
4001     qemu_opts_print_help(drv->amend_opts, false);
4002     return 0;
4003 }
4004 
4005 static int img_amend(int argc, char **argv)
4006 {
4007     Error *err = NULL;
4008     int c, ret = 0;
4009     char *options = NULL;
4010     QemuOptsList *amend_opts = NULL;
4011     QemuOpts *opts = NULL;
4012     const char *fmt = NULL, *filename, *cache;
4013     int flags;
4014     bool writethrough;
4015     bool quiet = false, progress = false;
4016     BlockBackend *blk = NULL;
4017     BlockDriverState *bs = NULL;
4018     bool image_opts = false;
4019     bool force = false;
4020 
4021     cache = BDRV_DEFAULT_CACHE;
4022     for (;;) {
4023         static const struct option long_options[] = {
4024             {"help", no_argument, 0, 'h'},
4025             {"object", required_argument, 0, OPTION_OBJECT},
4026             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4027             {"force", no_argument, 0, OPTION_FORCE},
4028             {0, 0, 0, 0}
4029         };
4030         c = getopt_long(argc, argv, ":ho:f:t:pq",
4031                         long_options, NULL);
4032         if (c == -1) {
4033             break;
4034         }
4035 
4036         switch (c) {
4037         case ':':
4038             missing_argument(argv[optind - 1]);
4039             break;
4040         case '?':
4041             unrecognized_option(argv[optind - 1]);
4042             break;
4043         case 'h':
4044             help();
4045             break;
4046         case 'o':
4047             if (accumulate_options(&options, optarg) < 0) {
4048                 ret = -1;
4049                 goto out_no_progress;
4050             }
4051             break;
4052         case 'f':
4053             fmt = optarg;
4054             break;
4055         case 't':
4056             cache = optarg;
4057             break;
4058         case 'p':
4059             progress = true;
4060             break;
4061         case 'q':
4062             quiet = true;
4063             break;
4064         case OPTION_OBJECT:
4065             user_creatable_process_cmdline(optarg);
4066             break;
4067         case OPTION_IMAGE_OPTS:
4068             image_opts = true;
4069             break;
4070         case OPTION_FORCE:
4071             force = true;
4072             break;
4073         }
4074     }
4075 
4076     if (!options) {
4077         error_exit("Must specify options (-o)");
4078     }
4079 
4080     if (quiet) {
4081         progress = false;
4082     }
4083     qemu_progress_init(progress, 1.0);
4084 
4085     filename = (optind == argc - 1) ? argv[argc - 1] : NULL;
4086     if (fmt && has_help_option(options)) {
4087         /* If a format is explicitly specified (and possibly no filename is
4088          * given), print option help here */
4089         ret = print_amend_option_help(fmt);
4090         goto out;
4091     }
4092 
4093     if (optind != argc - 1) {
4094         error_report("Expecting one image file name");
4095         ret = -1;
4096         goto out;
4097     }
4098 
4099     flags = BDRV_O_RDWR;
4100     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
4101     if (ret < 0) {
4102         error_report("Invalid cache option: %s", cache);
4103         goto out;
4104     }
4105 
4106     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
4107                    false);
4108     if (!blk) {
4109         ret = -1;
4110         goto out;
4111     }
4112     bs = blk_bs(blk);
4113 
4114     fmt = bs->drv->format_name;
4115 
4116     if (has_help_option(options)) {
4117         /* If the format was auto-detected, print option help here */
4118         ret = print_amend_option_help(fmt);
4119         goto out;
4120     }
4121 
4122     if (!bs->drv->bdrv_amend_options) {
4123         error_report("Format driver '%s' does not support option amendment",
4124                      fmt);
4125         ret = -1;
4126         goto out;
4127     }
4128 
4129     /* Every driver supporting amendment must have amend_opts */
4130     assert(bs->drv->amend_opts);
4131 
4132     amend_opts = qemu_opts_append(amend_opts, bs->drv->amend_opts);
4133     opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
4134     if (!qemu_opts_do_parse(opts, options, NULL, &err)) {
4135         /* Try to parse options using the create options */
4136         amend_opts = qemu_opts_append(amend_opts, bs->drv->create_opts);
4137         qemu_opts_del(opts);
4138         opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
4139         if (qemu_opts_do_parse(opts, options, NULL, NULL)) {
4140             error_append_hint(&err,
4141                               "This option is only supported for image creation\n");
4142         }
4143 
4144         error_report_err(err);
4145         ret = -1;
4146         goto out;
4147     }
4148 
4149     /* In case the driver does not call amend_status_cb() */
4150     qemu_progress_print(0.f, 0);
4151     ret = bdrv_amend_options(bs, opts, &amend_status_cb, NULL, force, &err);
4152     qemu_progress_print(100.f, 0);
4153     if (ret < 0) {
4154         error_report_err(err);
4155         goto out;
4156     }
4157 
4158 out:
4159     qemu_progress_end();
4160 
4161 out_no_progress:
4162     blk_unref(blk);
4163     qemu_opts_del(opts);
4164     qemu_opts_free(amend_opts);
4165     g_free(options);
4166 
4167     if (ret) {
4168         return 1;
4169     }
4170     return 0;
4171 }
4172 
4173 typedef struct BenchData {
4174     BlockBackend *blk;
4175     uint64_t image_size;
4176     bool write;
4177     int bufsize;
4178     int step;
4179     int nrreq;
4180     int n;
4181     int flush_interval;
4182     bool drain_on_flush;
4183     uint8_t *buf;
4184     QEMUIOVector *qiov;
4185 
4186     int in_flight;
4187     bool in_flush;
4188     uint64_t offset;
4189 } BenchData;
4190 
4191 static void bench_undrained_flush_cb(void *opaque, int ret)
4192 {
4193     if (ret < 0) {
4194         error_report("Failed flush request: %s", strerror(-ret));
4195         exit(EXIT_FAILURE);
4196     }
4197 }
4198 
4199 static void bench_cb(void *opaque, int ret)
4200 {
4201     BenchData *b = opaque;
4202     BlockAIOCB *acb;
4203 
4204     if (ret < 0) {
4205         error_report("Failed request: %s", strerror(-ret));
4206         exit(EXIT_FAILURE);
4207     }
4208 
4209     if (b->in_flush) {
4210         /* Just finished a flush with drained queue: Start next requests */
4211         assert(b->in_flight == 0);
4212         b->in_flush = false;
4213     } else if (b->in_flight > 0) {
4214         int remaining = b->n - b->in_flight;
4215 
4216         b->n--;
4217         b->in_flight--;
4218 
4219         /* Time for flush? Drain queue if requested, then flush */
4220         if (b->flush_interval && remaining % b->flush_interval == 0) {
4221             if (!b->in_flight || !b->drain_on_flush) {
4222                 BlockCompletionFunc *cb;
4223 
4224                 if (b->drain_on_flush) {
4225                     b->in_flush = true;
4226                     cb = bench_cb;
4227                 } else {
4228                     cb = bench_undrained_flush_cb;
4229                 }
4230 
4231                 acb = blk_aio_flush(b->blk, cb, b);
4232                 if (!acb) {
4233                     error_report("Failed to issue flush request");
4234                     exit(EXIT_FAILURE);
4235                 }
4236             }
4237             if (b->drain_on_flush) {
4238                 return;
4239             }
4240         }
4241     }
4242 
4243     while (b->n > b->in_flight && b->in_flight < b->nrreq) {
4244         int64_t offset = b->offset;
4245         /* blk_aio_* might look for completed I/Os and kick bench_cb
4246          * again, so make sure this operation is counted by in_flight
4247          * and b->offset is ready for the next submission.
4248          */
4249         b->in_flight++;
4250         b->offset += b->step;
4251         b->offset %= b->image_size;
4252         if (b->write) {
4253             acb = blk_aio_pwritev(b->blk, offset, b->qiov, 0, bench_cb, b);
4254         } else {
4255             acb = blk_aio_preadv(b->blk, offset, b->qiov, 0, bench_cb, b);
4256         }
4257         if (!acb) {
4258             error_report("Failed to issue request");
4259             exit(EXIT_FAILURE);
4260         }
4261     }
4262 }
4263 
4264 static int img_bench(int argc, char **argv)
4265 {
4266     int c, ret = 0;
4267     const char *fmt = NULL, *filename;
4268     bool quiet = false;
4269     bool image_opts = false;
4270     bool is_write = false;
4271     int count = 75000;
4272     int depth = 64;
4273     int64_t offset = 0;
4274     size_t bufsize = 4096;
4275     int pattern = 0;
4276     size_t step = 0;
4277     int flush_interval = 0;
4278     bool drain_on_flush = true;
4279     int64_t image_size;
4280     BlockBackend *blk = NULL;
4281     BenchData data = {};
4282     int flags = 0;
4283     bool writethrough = false;
4284     struct timeval t1, t2;
4285     int i;
4286     bool force_share = false;
4287     size_t buf_size;
4288 
4289     for (;;) {
4290         static const struct option long_options[] = {
4291             {"help", no_argument, 0, 'h'},
4292             {"flush-interval", required_argument, 0, OPTION_FLUSH_INTERVAL},
4293             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4294             {"pattern", required_argument, 0, OPTION_PATTERN},
4295             {"no-drain", no_argument, 0, OPTION_NO_DRAIN},
4296             {"force-share", no_argument, 0, 'U'},
4297             {0, 0, 0, 0}
4298         };
4299         c = getopt_long(argc, argv, ":hc:d:f:ni:o:qs:S:t:wU", long_options,
4300                         NULL);
4301         if (c == -1) {
4302             break;
4303         }
4304 
4305         switch (c) {
4306         case ':':
4307             missing_argument(argv[optind - 1]);
4308             break;
4309         case '?':
4310             unrecognized_option(argv[optind - 1]);
4311             break;
4312         case 'h':
4313             help();
4314             break;
4315         case 'c':
4316         {
4317             unsigned long res;
4318 
4319             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
4320                 error_report("Invalid request count specified");
4321                 return 1;
4322             }
4323             count = res;
4324             break;
4325         }
4326         case 'd':
4327         {
4328             unsigned long res;
4329 
4330             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
4331                 error_report("Invalid queue depth specified");
4332                 return 1;
4333             }
4334             depth = res;
4335             break;
4336         }
4337         case 'f':
4338             fmt = optarg;
4339             break;
4340         case 'n':
4341             flags |= BDRV_O_NATIVE_AIO;
4342             break;
4343         case 'i':
4344             ret = bdrv_parse_aio(optarg, &flags);
4345             if (ret < 0) {
4346                 error_report("Invalid aio option: %s", optarg);
4347                 ret = -1;
4348                 goto out;
4349             }
4350             break;
4351         case 'o':
4352         {
4353             offset = cvtnum("offset", optarg);
4354             if (offset < 0) {
4355                 return 1;
4356             }
4357             break;
4358         }
4359             break;
4360         case 'q':
4361             quiet = true;
4362             break;
4363         case 's':
4364         {
4365             int64_t sval;
4366 
4367             sval = cvtnum_full("buffer size", optarg, 0, INT_MAX);
4368             if (sval < 0) {
4369                 return 1;
4370             }
4371 
4372             bufsize = sval;
4373             break;
4374         }
4375         case 'S':
4376         {
4377             int64_t sval;
4378 
4379             sval = cvtnum_full("step_size", optarg, 0, INT_MAX);
4380             if (sval < 0) {
4381                 return 1;
4382             }
4383 
4384             step = sval;
4385             break;
4386         }
4387         case 't':
4388             ret = bdrv_parse_cache_mode(optarg, &flags, &writethrough);
4389             if (ret < 0) {
4390                 error_report("Invalid cache mode");
4391                 ret = -1;
4392                 goto out;
4393             }
4394             break;
4395         case 'w':
4396             flags |= BDRV_O_RDWR;
4397             is_write = true;
4398             break;
4399         case 'U':
4400             force_share = true;
4401             break;
4402         case OPTION_PATTERN:
4403         {
4404             unsigned long res;
4405 
4406             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > 0xff) {
4407                 error_report("Invalid pattern byte specified");
4408                 return 1;
4409             }
4410             pattern = res;
4411             break;
4412         }
4413         case OPTION_FLUSH_INTERVAL:
4414         {
4415             unsigned long res;
4416 
4417             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
4418                 error_report("Invalid flush interval specified");
4419                 return 1;
4420             }
4421             flush_interval = res;
4422             break;
4423         }
4424         case OPTION_NO_DRAIN:
4425             drain_on_flush = false;
4426             break;
4427         case OPTION_IMAGE_OPTS:
4428             image_opts = true;
4429             break;
4430         }
4431     }
4432 
4433     if (optind != argc - 1) {
4434         error_exit("Expecting one image file name");
4435     }
4436     filename = argv[argc - 1];
4437 
4438     if (!is_write && flush_interval) {
4439         error_report("--flush-interval is only available in write tests");
4440         ret = -1;
4441         goto out;
4442     }
4443     if (flush_interval && flush_interval < depth) {
4444         error_report("Flush interval can't be smaller than depth");
4445         ret = -1;
4446         goto out;
4447     }
4448 
4449     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
4450                    force_share);
4451     if (!blk) {
4452         ret = -1;
4453         goto out;
4454     }
4455 
4456     image_size = blk_getlength(blk);
4457     if (image_size < 0) {
4458         ret = image_size;
4459         goto out;
4460     }
4461 
4462     data = (BenchData) {
4463         .blk            = blk,
4464         .image_size     = image_size,
4465         .bufsize        = bufsize,
4466         .step           = step ?: bufsize,
4467         .nrreq          = depth,
4468         .n              = count,
4469         .offset         = offset,
4470         .write          = is_write,
4471         .flush_interval = flush_interval,
4472         .drain_on_flush = drain_on_flush,
4473     };
4474     printf("Sending %d %s requests, %d bytes each, %d in parallel "
4475            "(starting at offset %" PRId64 ", step size %d)\n",
4476            data.n, data.write ? "write" : "read", data.bufsize, data.nrreq,
4477            data.offset, data.step);
4478     if (flush_interval) {
4479         printf("Sending flush every %d requests\n", flush_interval);
4480     }
4481 
4482     buf_size = data.nrreq * data.bufsize;
4483     data.buf = blk_blockalign(blk, buf_size);
4484     memset(data.buf, pattern, data.nrreq * data.bufsize);
4485 
4486     blk_register_buf(blk, data.buf, buf_size);
4487 
4488     data.qiov = g_new(QEMUIOVector, data.nrreq);
4489     for (i = 0; i < data.nrreq; i++) {
4490         qemu_iovec_init(&data.qiov[i], 1);
4491         qemu_iovec_add(&data.qiov[i],
4492                        data.buf + i * data.bufsize, data.bufsize);
4493     }
4494 
4495     gettimeofday(&t1, NULL);
4496     bench_cb(&data, 0);
4497 
4498     while (data.n > 0) {
4499         main_loop_wait(false);
4500     }
4501     gettimeofday(&t2, NULL);
4502 
4503     printf("Run completed in %3.3f seconds.\n",
4504            (t2.tv_sec - t1.tv_sec)
4505            + ((double)(t2.tv_usec - t1.tv_usec) / 1000000));
4506 
4507 out:
4508     if (data.buf) {
4509         blk_unregister_buf(blk, data.buf);
4510     }
4511     qemu_vfree(data.buf);
4512     blk_unref(blk);
4513 
4514     if (ret) {
4515         return 1;
4516     }
4517     return 0;
4518 }
4519 
4520 enum ImgBitmapAct {
4521     BITMAP_ADD,
4522     BITMAP_REMOVE,
4523     BITMAP_CLEAR,
4524     BITMAP_ENABLE,
4525     BITMAP_DISABLE,
4526     BITMAP_MERGE,
4527 };
4528 typedef struct ImgBitmapAction {
4529     enum ImgBitmapAct act;
4530     const char *src; /* only used for merge */
4531     QSIMPLEQ_ENTRY(ImgBitmapAction) next;
4532 } ImgBitmapAction;
4533 
4534 static int img_bitmap(int argc, char **argv)
4535 {
4536     Error *err = NULL;
4537     int c, ret = 1;
4538     QemuOpts *opts = NULL;
4539     const char *fmt = NULL, *src_fmt = NULL, *src_filename = NULL;
4540     const char *filename, *bitmap;
4541     BlockBackend *blk = NULL, *src = NULL;
4542     BlockDriverState *bs = NULL, *src_bs = NULL;
4543     bool image_opts = false;
4544     int64_t granularity = 0;
4545     bool add = false, merge = false;
4546     QSIMPLEQ_HEAD(, ImgBitmapAction) actions;
4547     ImgBitmapAction *act, *act_next;
4548     const char *op;
4549 
4550     QSIMPLEQ_INIT(&actions);
4551 
4552     for (;;) {
4553         static const struct option long_options[] = {
4554             {"help", no_argument, 0, 'h'},
4555             {"object", required_argument, 0, OPTION_OBJECT},
4556             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4557             {"add", no_argument, 0, OPTION_ADD},
4558             {"remove", no_argument, 0, OPTION_REMOVE},
4559             {"clear", no_argument, 0, OPTION_CLEAR},
4560             {"enable", no_argument, 0, OPTION_ENABLE},
4561             {"disable", no_argument, 0, OPTION_DISABLE},
4562             {"merge", required_argument, 0, OPTION_MERGE},
4563             {"granularity", required_argument, 0, 'g'},
4564             {"source-file", required_argument, 0, 'b'},
4565             {"source-format", required_argument, 0, 'F'},
4566             {0, 0, 0, 0}
4567         };
4568         c = getopt_long(argc, argv, ":b:f:F:g:h", long_options, NULL);
4569         if (c == -1) {
4570             break;
4571         }
4572 
4573         switch (c) {
4574         case ':':
4575             missing_argument(argv[optind - 1]);
4576             break;
4577         case '?':
4578             unrecognized_option(argv[optind - 1]);
4579             break;
4580         case 'h':
4581             help();
4582             break;
4583         case 'b':
4584             src_filename = optarg;
4585             break;
4586         case 'f':
4587             fmt = optarg;
4588             break;
4589         case 'F':
4590             src_fmt = optarg;
4591             break;
4592         case 'g':
4593             granularity = cvtnum("granularity", optarg);
4594             if (granularity < 0) {
4595                 return 1;
4596             }
4597             break;
4598         case OPTION_ADD:
4599             act = g_new0(ImgBitmapAction, 1);
4600             act->act = BITMAP_ADD;
4601             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4602             add = true;
4603             break;
4604         case OPTION_REMOVE:
4605             act = g_new0(ImgBitmapAction, 1);
4606             act->act = BITMAP_REMOVE;
4607             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4608             break;
4609         case OPTION_CLEAR:
4610             act = g_new0(ImgBitmapAction, 1);
4611             act->act = BITMAP_CLEAR;
4612             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4613             break;
4614         case OPTION_ENABLE:
4615             act = g_new0(ImgBitmapAction, 1);
4616             act->act = BITMAP_ENABLE;
4617             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4618             break;
4619         case OPTION_DISABLE:
4620             act = g_new0(ImgBitmapAction, 1);
4621             act->act = BITMAP_DISABLE;
4622             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4623             break;
4624         case OPTION_MERGE:
4625             act = g_new0(ImgBitmapAction, 1);
4626             act->act = BITMAP_MERGE;
4627             act->src = optarg;
4628             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4629             merge = true;
4630             break;
4631         case OPTION_OBJECT:
4632             user_creatable_process_cmdline(optarg);
4633             break;
4634         case OPTION_IMAGE_OPTS:
4635             image_opts = true;
4636             break;
4637         }
4638     }
4639 
4640     if (QSIMPLEQ_EMPTY(&actions)) {
4641         error_report("Need at least one of --add, --remove, --clear, "
4642                      "--enable, --disable, or --merge");
4643         goto out;
4644     }
4645 
4646     if (granularity && !add) {
4647         error_report("granularity only supported with --add");
4648         goto out;
4649     }
4650     if (src_fmt && !src_filename) {
4651         error_report("-F only supported with -b");
4652         goto out;
4653     }
4654     if (src_filename && !merge) {
4655         error_report("Merge bitmap source file only supported with "
4656                      "--merge");
4657         goto out;
4658     }
4659 
4660     if (optind != argc - 2) {
4661         error_report("Expecting filename and bitmap name");
4662         goto out;
4663     }
4664 
4665     filename = argv[optind];
4666     bitmap = argv[optind + 1];
4667 
4668     /*
4669      * No need to open backing chains; we will be manipulating bitmaps
4670      * directly in this image without reference to image contents.
4671      */
4672     blk = img_open(image_opts, filename, fmt, BDRV_O_RDWR | BDRV_O_NO_BACKING,
4673                    false, false, false);
4674     if (!blk) {
4675         goto out;
4676     }
4677     bs = blk_bs(blk);
4678     if (src_filename) {
4679         src = img_open(false, src_filename, src_fmt, BDRV_O_NO_BACKING,
4680                        false, false, false);
4681         if (!src) {
4682             goto out;
4683         }
4684         src_bs = blk_bs(src);
4685     } else {
4686         src_bs = bs;
4687     }
4688 
4689     QSIMPLEQ_FOREACH_SAFE(act, &actions, next, act_next) {
4690         switch (act->act) {
4691         case BITMAP_ADD:
4692             qmp_block_dirty_bitmap_add(bs->node_name, bitmap,
4693                                        !!granularity, granularity, true, true,
4694                                        false, false, &err);
4695             op = "add";
4696             break;
4697         case BITMAP_REMOVE:
4698             qmp_block_dirty_bitmap_remove(bs->node_name, bitmap, &err);
4699             op = "remove";
4700             break;
4701         case BITMAP_CLEAR:
4702             qmp_block_dirty_bitmap_clear(bs->node_name, bitmap, &err);
4703             op = "clear";
4704             break;
4705         case BITMAP_ENABLE:
4706             qmp_block_dirty_bitmap_enable(bs->node_name, bitmap, &err);
4707             op = "enable";
4708             break;
4709         case BITMAP_DISABLE:
4710             qmp_block_dirty_bitmap_disable(bs->node_name, bitmap, &err);
4711             op = "disable";
4712             break;
4713         case BITMAP_MERGE:
4714             do_dirty_bitmap_merge(bs->node_name, bitmap, src_bs->node_name,
4715                                   act->src, &err);
4716             op = "merge";
4717             break;
4718         default:
4719             g_assert_not_reached();
4720         }
4721 
4722         if (err) {
4723             error_reportf_err(err, "Operation %s on bitmap %s failed: ",
4724                               op, bitmap);
4725             goto out;
4726         }
4727         g_free(act);
4728     }
4729 
4730     ret = 0;
4731 
4732  out:
4733     blk_unref(src);
4734     blk_unref(blk);
4735     qemu_opts_del(opts);
4736     return ret;
4737 }
4738 
4739 #define C_BS      01
4740 #define C_COUNT   02
4741 #define C_IF      04
4742 #define C_OF      010
4743 #define C_SKIP    020
4744 
4745 struct DdInfo {
4746     unsigned int flags;
4747     int64_t count;
4748 };
4749 
4750 struct DdIo {
4751     int bsz;    /* Block size */
4752     char *filename;
4753     uint8_t *buf;
4754     int64_t offset;
4755 };
4756 
4757 struct DdOpts {
4758     const char *name;
4759     int (*f)(const char *, struct DdIo *, struct DdIo *, struct DdInfo *);
4760     unsigned int flag;
4761 };
4762 
4763 static int img_dd_bs(const char *arg,
4764                      struct DdIo *in, struct DdIo *out,
4765                      struct DdInfo *dd)
4766 {
4767     int64_t res;
4768 
4769     res = cvtnum_full("bs", arg, 1, INT_MAX);
4770 
4771     if (res < 0) {
4772         return 1;
4773     }
4774     in->bsz = out->bsz = res;
4775 
4776     return 0;
4777 }
4778 
4779 static int img_dd_count(const char *arg,
4780                         struct DdIo *in, struct DdIo *out,
4781                         struct DdInfo *dd)
4782 {
4783     dd->count = cvtnum("count", arg);
4784 
4785     if (dd->count < 0) {
4786         return 1;
4787     }
4788 
4789     return 0;
4790 }
4791 
4792 static int img_dd_if(const char *arg,
4793                      struct DdIo *in, struct DdIo *out,
4794                      struct DdInfo *dd)
4795 {
4796     in->filename = g_strdup(arg);
4797 
4798     return 0;
4799 }
4800 
4801 static int img_dd_of(const char *arg,
4802                      struct DdIo *in, struct DdIo *out,
4803                      struct DdInfo *dd)
4804 {
4805     out->filename = g_strdup(arg);
4806 
4807     return 0;
4808 }
4809 
4810 static int img_dd_skip(const char *arg,
4811                        struct DdIo *in, struct DdIo *out,
4812                        struct DdInfo *dd)
4813 {
4814     in->offset = cvtnum("skip", arg);
4815 
4816     if (in->offset < 0) {
4817         return 1;
4818     }
4819 
4820     return 0;
4821 }
4822 
4823 static int img_dd(int argc, char **argv)
4824 {
4825     int ret = 0;
4826     char *arg = NULL;
4827     char *tmp;
4828     BlockDriver *drv = NULL, *proto_drv = NULL;
4829     BlockBackend *blk1 = NULL, *blk2 = NULL;
4830     QemuOpts *opts = NULL;
4831     QemuOptsList *create_opts = NULL;
4832     Error *local_err = NULL;
4833     bool image_opts = false;
4834     int c, i;
4835     const char *out_fmt = "raw";
4836     const char *fmt = NULL;
4837     int64_t size = 0;
4838     int64_t block_count = 0, out_pos, in_pos;
4839     bool force_share = false;
4840     struct DdInfo dd = {
4841         .flags = 0,
4842         .count = 0,
4843     };
4844     struct DdIo in = {
4845         .bsz = 512, /* Block size is by default 512 bytes */
4846         .filename = NULL,
4847         .buf = NULL,
4848         .offset = 0
4849     };
4850     struct DdIo out = {
4851         .bsz = 512,
4852         .filename = NULL,
4853         .buf = NULL,
4854         .offset = 0
4855     };
4856 
4857     const struct DdOpts options[] = {
4858         { "bs", img_dd_bs, C_BS },
4859         { "count", img_dd_count, C_COUNT },
4860         { "if", img_dd_if, C_IF },
4861         { "of", img_dd_of, C_OF },
4862         { "skip", img_dd_skip, C_SKIP },
4863         { NULL, NULL, 0 }
4864     };
4865     const struct option long_options[] = {
4866         { "help", no_argument, 0, 'h'},
4867         { "object", required_argument, 0, OPTION_OBJECT},
4868         { "image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4869         { "force-share", no_argument, 0, 'U'},
4870         { 0, 0, 0, 0 }
4871     };
4872 
4873     while ((c = getopt_long(argc, argv, ":hf:O:U", long_options, NULL))) {
4874         if (c == EOF) {
4875             break;
4876         }
4877         switch (c) {
4878         case 'O':
4879             out_fmt = optarg;
4880             break;
4881         case 'f':
4882             fmt = optarg;
4883             break;
4884         case ':':
4885             missing_argument(argv[optind - 1]);
4886             break;
4887         case '?':
4888             unrecognized_option(argv[optind - 1]);
4889             break;
4890         case 'h':
4891             help();
4892             break;
4893         case 'U':
4894             force_share = true;
4895             break;
4896         case OPTION_OBJECT:
4897             user_creatable_process_cmdline(optarg);
4898             break;
4899         case OPTION_IMAGE_OPTS:
4900             image_opts = true;
4901             break;
4902         }
4903     }
4904 
4905     for (i = optind; i < argc; i++) {
4906         int j;
4907         arg = g_strdup(argv[i]);
4908 
4909         tmp = strchr(arg, '=');
4910         if (tmp == NULL) {
4911             error_report("unrecognized operand %s", arg);
4912             ret = -1;
4913             goto out;
4914         }
4915 
4916         *tmp++ = '\0';
4917 
4918         for (j = 0; options[j].name != NULL; j++) {
4919             if (!strcmp(arg, options[j].name)) {
4920                 break;
4921             }
4922         }
4923         if (options[j].name == NULL) {
4924             error_report("unrecognized operand %s", arg);
4925             ret = -1;
4926             goto out;
4927         }
4928 
4929         if (options[j].f(tmp, &in, &out, &dd) != 0) {
4930             ret = -1;
4931             goto out;
4932         }
4933         dd.flags |= options[j].flag;
4934         g_free(arg);
4935         arg = NULL;
4936     }
4937 
4938     if (!(dd.flags & C_IF && dd.flags & C_OF)) {
4939         error_report("Must specify both input and output files");
4940         ret = -1;
4941         goto out;
4942     }
4943 
4944     blk1 = img_open(image_opts, in.filename, fmt, 0, false, false,
4945                     force_share);
4946 
4947     if (!blk1) {
4948         ret = -1;
4949         goto out;
4950     }
4951 
4952     drv = bdrv_find_format(out_fmt);
4953     if (!drv) {
4954         error_report("Unknown file format");
4955         ret = -1;
4956         goto out;
4957     }
4958     proto_drv = bdrv_find_protocol(out.filename, true, &local_err);
4959 
4960     if (!proto_drv) {
4961         error_report_err(local_err);
4962         ret = -1;
4963         goto out;
4964     }
4965     if (!drv->create_opts) {
4966         error_report("Format driver '%s' does not support image creation",
4967                      drv->format_name);
4968         ret = -1;
4969         goto out;
4970     }
4971     if (!proto_drv->create_opts) {
4972         error_report("Protocol driver '%s' does not support image creation",
4973                      proto_drv->format_name);
4974         ret = -1;
4975         goto out;
4976     }
4977     create_opts = qemu_opts_append(create_opts, drv->create_opts);
4978     create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4979 
4980     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4981 
4982     size = blk_getlength(blk1);
4983     if (size < 0) {
4984         error_report("Failed to get size for '%s'", in.filename);
4985         ret = -1;
4986         goto out;
4987     }
4988 
4989     if (dd.flags & C_COUNT && dd.count <= INT64_MAX / in.bsz &&
4990         dd.count * in.bsz < size) {
4991         size = dd.count * in.bsz;
4992     }
4993 
4994     /* Overflow means the specified offset is beyond input image's size */
4995     if (dd.flags & C_SKIP && (in.offset > INT64_MAX / in.bsz ||
4996                               size < in.bsz * in.offset)) {
4997         qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 0, &error_abort);
4998     } else {
4999         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
5000                             size - in.bsz * in.offset, &error_abort);
5001     }
5002 
5003     ret = bdrv_create(drv, out.filename, opts, &local_err);
5004     if (ret < 0) {
5005         error_reportf_err(local_err,
5006                           "%s: error while creating output image: ",
5007                           out.filename);
5008         ret = -1;
5009         goto out;
5010     }
5011 
5012     /* TODO, we can't honour --image-opts for the target,
5013      * since it needs to be given in a format compatible
5014      * with the bdrv_create() call above which does not
5015      * support image-opts style.
5016      */
5017     blk2 = img_open_file(out.filename, NULL, out_fmt, BDRV_O_RDWR,
5018                          false, false, false);
5019 
5020     if (!blk2) {
5021         ret = -1;
5022         goto out;
5023     }
5024 
5025     if (dd.flags & C_SKIP && (in.offset > INT64_MAX / in.bsz ||
5026                               size < in.offset * in.bsz)) {
5027         /* We give a warning if the skip option is bigger than the input
5028          * size and create an empty output disk image (i.e. like dd(1)).
5029          */
5030         error_report("%s: cannot skip to specified offset", in.filename);
5031         in_pos = size;
5032     } else {
5033         in_pos = in.offset * in.bsz;
5034     }
5035 
5036     in.buf = g_new(uint8_t, in.bsz);
5037 
5038     for (out_pos = 0; in_pos < size; block_count++) {
5039         int in_ret, out_ret;
5040 
5041         if (in_pos + in.bsz > size) {
5042             in_ret = blk_pread(blk1, in_pos, in.buf, size - in_pos);
5043         } else {
5044             in_ret = blk_pread(blk1, in_pos, in.buf, in.bsz);
5045         }
5046         if (in_ret < 0) {
5047             error_report("error while reading from input image file: %s",
5048                          strerror(-in_ret));
5049             ret = -1;
5050             goto out;
5051         }
5052         in_pos += in_ret;
5053 
5054         out_ret = blk_pwrite(blk2, out_pos, in.buf, in_ret, 0);
5055 
5056         if (out_ret < 0) {
5057             error_report("error while writing to output image file: %s",
5058                          strerror(-out_ret));
5059             ret = -1;
5060             goto out;
5061         }
5062         out_pos += out_ret;
5063     }
5064 
5065 out:
5066     g_free(arg);
5067     qemu_opts_del(opts);
5068     qemu_opts_free(create_opts);
5069     blk_unref(blk1);
5070     blk_unref(blk2);
5071     g_free(in.filename);
5072     g_free(out.filename);
5073     g_free(in.buf);
5074     g_free(out.buf);
5075 
5076     if (ret) {
5077         return 1;
5078     }
5079     return 0;
5080 }
5081 
5082 static void dump_json_block_measure_info(BlockMeasureInfo *info)
5083 {
5084     GString *str;
5085     QObject *obj;
5086     Visitor *v = qobject_output_visitor_new(&obj);
5087 
5088     visit_type_BlockMeasureInfo(v, NULL, &info, &error_abort);
5089     visit_complete(v, &obj);
5090     str = qobject_to_json_pretty(obj, true);
5091     assert(str != NULL);
5092     printf("%s\n", str->str);
5093     qobject_unref(obj);
5094     visit_free(v);
5095     g_string_free(str, true);
5096 }
5097 
5098 static int img_measure(int argc, char **argv)
5099 {
5100     static const struct option long_options[] = {
5101         {"help", no_argument, 0, 'h'},
5102         {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
5103         {"object", required_argument, 0, OPTION_OBJECT},
5104         {"output", required_argument, 0, OPTION_OUTPUT},
5105         {"size", required_argument, 0, OPTION_SIZE},
5106         {"force-share", no_argument, 0, 'U'},
5107         {0, 0, 0, 0}
5108     };
5109     OutputFormat output_format = OFORMAT_HUMAN;
5110     BlockBackend *in_blk = NULL;
5111     BlockDriver *drv;
5112     const char *filename = NULL;
5113     const char *fmt = NULL;
5114     const char *out_fmt = "raw";
5115     char *options = NULL;
5116     char *snapshot_name = NULL;
5117     bool force_share = false;
5118     QemuOpts *opts = NULL;
5119     QemuOpts *object_opts = NULL;
5120     QemuOpts *sn_opts = NULL;
5121     QemuOptsList *create_opts = NULL;
5122     bool image_opts = false;
5123     uint64_t img_size = UINT64_MAX;
5124     BlockMeasureInfo *info = NULL;
5125     Error *local_err = NULL;
5126     int ret = 1;
5127     int c;
5128 
5129     while ((c = getopt_long(argc, argv, "hf:O:o:l:U",
5130                             long_options, NULL)) != -1) {
5131         switch (c) {
5132         case '?':
5133         case 'h':
5134             help();
5135             break;
5136         case 'f':
5137             fmt = optarg;
5138             break;
5139         case 'O':
5140             out_fmt = optarg;
5141             break;
5142         case 'o':
5143             if (accumulate_options(&options, optarg) < 0) {
5144                 goto out;
5145             }
5146             break;
5147         case 'l':
5148             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
5149                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
5150                                                   optarg, false);
5151                 if (!sn_opts) {
5152                     error_report("Failed in parsing snapshot param '%s'",
5153                                  optarg);
5154                     goto out;
5155                 }
5156             } else {
5157                 snapshot_name = optarg;
5158             }
5159             break;
5160         case 'U':
5161             force_share = true;
5162             break;
5163         case OPTION_OBJECT:
5164             user_creatable_process_cmdline(optarg);
5165             break;
5166         case OPTION_IMAGE_OPTS:
5167             image_opts = true;
5168             break;
5169         case OPTION_OUTPUT:
5170             if (!strcmp(optarg, "json")) {
5171                 output_format = OFORMAT_JSON;
5172             } else if (!strcmp(optarg, "human")) {
5173                 output_format = OFORMAT_HUMAN;
5174             } else {
5175                 error_report("--output must be used with human or json "
5176                              "as argument.");
5177                 goto out;
5178             }
5179             break;
5180         case OPTION_SIZE:
5181         {
5182             int64_t sval;
5183 
5184             sval = cvtnum("image size", optarg);
5185             if (sval < 0) {
5186                 goto out;
5187             }
5188             img_size = (uint64_t)sval;
5189         }
5190         break;
5191         }
5192     }
5193 
5194     if (argc - optind > 1) {
5195         error_report("At most one filename argument is allowed.");
5196         goto out;
5197     } else if (argc - optind == 1) {
5198         filename = argv[optind];
5199     }
5200 
5201     if (!filename && (image_opts || fmt || snapshot_name || sn_opts)) {
5202         error_report("--image-opts, -f, and -l require a filename argument.");
5203         goto out;
5204     }
5205     if (filename && img_size != UINT64_MAX) {
5206         error_report("--size N cannot be used together with a filename.");
5207         goto out;
5208     }
5209     if (!filename && img_size == UINT64_MAX) {
5210         error_report("Either --size N or one filename must be specified.");
5211         goto out;
5212     }
5213 
5214     if (filename) {
5215         in_blk = img_open(image_opts, filename, fmt, 0,
5216                           false, false, force_share);
5217         if (!in_blk) {
5218             goto out;
5219         }
5220 
5221         if (sn_opts) {
5222             bdrv_snapshot_load_tmp(blk_bs(in_blk),
5223                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
5224                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
5225                     &local_err);
5226         } else if (snapshot_name != NULL) {
5227             bdrv_snapshot_load_tmp_by_id_or_name(blk_bs(in_blk),
5228                     snapshot_name, &local_err);
5229         }
5230         if (local_err) {
5231             error_reportf_err(local_err, "Failed to load snapshot: ");
5232             goto out;
5233         }
5234     }
5235 
5236     drv = bdrv_find_format(out_fmt);
5237     if (!drv) {
5238         error_report("Unknown file format '%s'", out_fmt);
5239         goto out;
5240     }
5241     if (!drv->create_opts) {
5242         error_report("Format driver '%s' does not support image creation",
5243                      drv->format_name);
5244         goto out;
5245     }
5246 
5247     create_opts = qemu_opts_append(create_opts, drv->create_opts);
5248     create_opts = qemu_opts_append(create_opts, bdrv_file.create_opts);
5249     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5250     if (options) {
5251         if (!qemu_opts_do_parse(opts, options, NULL, &local_err)) {
5252             error_report_err(local_err);
5253             error_report("Invalid options for file format '%s'", out_fmt);
5254             goto out;
5255         }
5256     }
5257     if (img_size != UINT64_MAX) {
5258         qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
5259     }
5260 
5261     info = bdrv_measure(drv, opts, in_blk ? blk_bs(in_blk) : NULL, &local_err);
5262     if (local_err) {
5263         error_report_err(local_err);
5264         goto out;
5265     }
5266 
5267     if (output_format == OFORMAT_HUMAN) {
5268         printf("required size: %" PRIu64 "\n", info->required);
5269         printf("fully allocated size: %" PRIu64 "\n", info->fully_allocated);
5270         if (info->has_bitmaps) {
5271             printf("bitmaps size: %" PRIu64 "\n", info->bitmaps);
5272         }
5273     } else {
5274         dump_json_block_measure_info(info);
5275     }
5276 
5277     ret = 0;
5278 
5279 out:
5280     qapi_free_BlockMeasureInfo(info);
5281     qemu_opts_del(object_opts);
5282     qemu_opts_del(opts);
5283     qemu_opts_del(sn_opts);
5284     qemu_opts_free(create_opts);
5285     g_free(options);
5286     blk_unref(in_blk);
5287     return ret;
5288 }
5289 
5290 static const img_cmd_t img_cmds[] = {
5291 #define DEF(option, callback, arg_string)        \
5292     { option, callback },
5293 #include "qemu-img-cmds.h"
5294 #undef DEF
5295     { NULL, NULL, },
5296 };
5297 
5298 int main(int argc, char **argv)
5299 {
5300     const img_cmd_t *cmd;
5301     const char *cmdname;
5302     Error *local_error = NULL;
5303     int c;
5304     static const struct option long_options[] = {
5305         {"help", no_argument, 0, 'h'},
5306         {"version", no_argument, 0, 'V'},
5307         {"trace", required_argument, NULL, 'T'},
5308         {0, 0, 0, 0}
5309     };
5310 
5311 #ifdef CONFIG_POSIX
5312     signal(SIGPIPE, SIG_IGN);
5313 #endif
5314 
5315     socket_init();
5316     error_init(argv[0]);
5317     module_call_init(MODULE_INIT_TRACE);
5318     qemu_init_exec_dir(argv[0]);
5319 
5320     if (qemu_init_main_loop(&local_error)) {
5321         error_report_err(local_error);
5322         exit(EXIT_FAILURE);
5323     }
5324 
5325     qcrypto_init(&error_fatal);
5326 
5327     module_call_init(MODULE_INIT_QOM);
5328     bdrv_init();
5329     if (argc < 2) {
5330         error_exit("Not enough arguments");
5331     }
5332 
5333     qemu_add_opts(&qemu_source_opts);
5334     qemu_add_opts(&qemu_trace_opts);
5335 
5336     while ((c = getopt_long(argc, argv, "+:hVT:", long_options, NULL)) != -1) {
5337         switch (c) {
5338         case ':':
5339             missing_argument(argv[optind - 1]);
5340             return 0;
5341         case '?':
5342             unrecognized_option(argv[optind - 1]);
5343             return 0;
5344         case 'h':
5345             help();
5346             return 0;
5347         case 'V':
5348             printf(QEMU_IMG_VERSION);
5349             return 0;
5350         case 'T':
5351             trace_opt_parse(optarg);
5352             break;
5353         }
5354     }
5355 
5356     cmdname = argv[optind];
5357 
5358     /* reset getopt_long scanning */
5359     argc -= optind;
5360     if (argc < 1) {
5361         return 0;
5362     }
5363     argv += optind;
5364     qemu_reset_optind();
5365 
5366     if (!trace_init_backends()) {
5367         exit(1);
5368     }
5369     trace_init_file();
5370     qemu_set_log(LOG_TRACE);
5371 
5372     /* find the command */
5373     for (cmd = img_cmds; cmd->name != NULL; cmd++) {
5374         if (!strcmp(cmdname, cmd->name)) {
5375             return cmd->handler(argc, argv);
5376         }
5377     }
5378 
5379     /* not found */
5380     error_exit("Command not found: %s", cmdname);
5381 }
5382