xref: /qemu/qemu-img.c (revision 29b62a10)
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/help-texts.h"
29 #include "qemu/qemu-progress.h"
30 #include "qemu-version.h"
31 #include "qapi/error.h"
32 #include "qapi/qapi-commands-block-core.h"
33 #include "qapi/qapi-visit-block-core.h"
34 #include "qapi/qobject-output-visitor.h"
35 #include "qapi/qmp/qjson.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qemu/cutils.h"
38 #include "qemu/config-file.h"
39 #include "qemu/option.h"
40 #include "qemu/error-report.h"
41 #include "qemu/log.h"
42 #include "qemu/main-loop.h"
43 #include "qemu/module.h"
44 #include "qemu/sockets.h"
45 #include "qemu/units.h"
46 #include "qemu/memalign.h"
47 #include "qom/object_interfaces.h"
48 #include "sysemu/block-backend.h"
49 #include "block/block_int.h"
50 #include "block/blockjob.h"
51 #include "block/dirty-bitmap.h"
52 #include "block/qapi.h"
53 #include "crypto/init.h"
54 #include "trace/control.h"
55 #include "qemu/throttle.h"
56 #include "block/throttle-groups.h"
57 
58 #define QEMU_IMG_VERSION "qemu-img version " QEMU_FULL_VERSION \
59                           "\n" QEMU_COPYRIGHT "\n"
60 
61 typedef struct img_cmd_t {
62     const char *name;
63     int (*handler)(int argc, char **argv);
64 } img_cmd_t;
65 
66 enum {
67     OPTION_OUTPUT = 256,
68     OPTION_BACKING_CHAIN = 257,
69     OPTION_OBJECT = 258,
70     OPTION_IMAGE_OPTS = 259,
71     OPTION_PATTERN = 260,
72     OPTION_FLUSH_INTERVAL = 261,
73     OPTION_NO_DRAIN = 262,
74     OPTION_TARGET_IMAGE_OPTS = 263,
75     OPTION_SIZE = 264,
76     OPTION_PREALLOCATION = 265,
77     OPTION_SHRINK = 266,
78     OPTION_SALVAGE = 267,
79     OPTION_TARGET_IS_ZERO = 268,
80     OPTION_ADD = 269,
81     OPTION_REMOVE = 270,
82     OPTION_CLEAR = 271,
83     OPTION_ENABLE = 272,
84     OPTION_DISABLE = 273,
85     OPTION_MERGE = 274,
86     OPTION_BITMAPS = 275,
87     OPTION_FORCE = 276,
88     OPTION_SKIP_BROKEN = 277,
89 };
90 
91 typedef enum OutputFormat {
92     OFORMAT_JSON,
93     OFORMAT_HUMAN,
94 } OutputFormat;
95 
96 /* Default to cache=writeback as data integrity is not important for qemu-img */
97 #define BDRV_DEFAULT_CACHE "writeback"
98 
99 static void format_print(void *opaque, const char *name)
100 {
101     printf(" %s", name);
102 }
103 
104 static G_NORETURN G_GNUC_PRINTF(1, 2)
105 void error_exit(const char *fmt, ...)
106 {
107     va_list ap;
108 
109     va_start(ap, fmt);
110     error_vreport(fmt, ap);
111     va_end(ap);
112 
113     error_printf("Try 'qemu-img --help' for more information\n");
114     exit(EXIT_FAILURE);
115 }
116 
117 static G_NORETURN
118 void missing_argument(const char *option)
119 {
120     error_exit("missing argument for option '%s'", option);
121 }
122 
123 static G_NORETURN
124 void unrecognized_option(const char *option)
125 {
126     error_exit("unrecognized option '%s'", option);
127 }
128 
129 /* Please keep in synch with docs/tools/qemu-img.rst */
130 static G_NORETURN
131 void help(void)
132 {
133     const char *help_msg =
134            QEMU_IMG_VERSION
135            "usage: qemu-img [standard options] command [command options]\n"
136            "QEMU disk image utility\n"
137            "\n"
138            "    '-h', '--help'       display this help and exit\n"
139            "    '-V', '--version'    output version information and exit\n"
140            "    '-T', '--trace'      [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
141            "                         specify tracing options\n"
142            "\n"
143            "Command syntax:\n"
144 #define DEF(option, callback, arg_string)        \
145            "  " arg_string "\n"
146 #include "qemu-img-cmds.h"
147 #undef DEF
148            "\n"
149            "Command parameters:\n"
150            "  'filename' is a disk image filename\n"
151            "  'objectdef' is a QEMU user creatable object definition. See the qemu(1)\n"
152            "    manual page for a description of the object properties. The most common\n"
153            "    object type is a 'secret', which is used to supply passwords and/or\n"
154            "    encryption keys.\n"
155            "  'fmt' is the disk image format. It is guessed automatically in most cases\n"
156            "  'cache' is the cache mode used to write the output disk image, the valid\n"
157            "    options are: 'none', 'writeback' (default, except for convert), 'writethrough',\n"
158            "    'directsync' and 'unsafe' (default for convert)\n"
159            "  'src_cache' is the cache mode used to read input disk images, the valid\n"
160            "    options are the same as for the 'cache' option\n"
161            "  'size' is the disk image size in bytes. Optional suffixes\n"
162            "    'k' or 'K' (kilobyte, 1024), 'M' (megabyte, 1024k), 'G' (gigabyte, 1024M),\n"
163            "    'T' (terabyte, 1024G), 'P' (petabyte, 1024T) and 'E' (exabyte, 1024P)  are\n"
164            "    supported. 'b' is ignored.\n"
165            "  'output_filename' is the destination disk image filename\n"
166            "  'output_fmt' is the destination format\n"
167            "  'options' is a comma separated list of format specific options in a\n"
168            "    name=value format. Use -o help for an overview of the options supported by\n"
169            "    the used format\n"
170            "  'snapshot_param' is param used for internal snapshot, format\n"
171            "    is 'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
172            "    '[ID_OR_NAME]'\n"
173            "  '-c' indicates that target image must be compressed (qcow format only)\n"
174            "  '-u' allows unsafe backing chains. For rebasing, it is assumed that old and\n"
175            "       new backing file match exactly. The image doesn't need a working\n"
176            "       backing file before rebasing in this case (useful for renaming the\n"
177            "       backing file). For image creation, allow creating without attempting\n"
178            "       to open the backing file.\n"
179            "  '-h' with or without a command shows this help and lists the supported formats\n"
180            "  '-p' show progress of command (only certain commands)\n"
181            "  '-q' use Quiet mode - do not print any output (except errors)\n"
182            "  '-S' indicates the consecutive number of bytes (defaults to 4k) that must\n"
183            "       contain only zeros for qemu-img to create a sparse image during\n"
184            "       conversion. If the number of bytes is 0, the source will not be scanned for\n"
185            "       unallocated or zero sectors, and the destination image will always be\n"
186            "       fully allocated\n"
187            "  '--output' takes the format in which the output must be done (human or json)\n"
188            "  '-n' skips the target volume creation (useful if the volume is created\n"
189            "       prior to running qemu-img)\n"
190            "\n"
191            "Parameters to bitmap subcommand:\n"
192            "  'bitmap' is the name of the bitmap to manipulate, through one or more\n"
193            "       actions from '--add', '--remove', '--clear', '--enable', '--disable',\n"
194            "       or '--merge source'\n"
195            "  '-g granularity' sets the granularity for '--add' actions\n"
196            "  '-b source' and '-F src_fmt' tell '--merge' actions to find the source\n"
197            "       bitmaps from an alternative file\n"
198            "\n"
199            "Parameters to check subcommand:\n"
200            "  '-r' tries to repair any inconsistencies that are found during the check.\n"
201            "       '-r leaks' repairs only cluster leaks, whereas '-r all' fixes all\n"
202            "       kinds of errors, with a higher risk of choosing the wrong fix or\n"
203            "       hiding corruption that has already occurred.\n"
204            "\n"
205            "Parameters to convert subcommand:\n"
206            "  '--bitmaps' copies all top-level persistent bitmaps to destination\n"
207            "  '-m' specifies how many coroutines work in parallel during the convert\n"
208            "       process (defaults to 8)\n"
209            "  '-W' allow to write to the target out of order rather than sequential\n"
210            "\n"
211            "Parameters to snapshot subcommand:\n"
212            "  'snapshot' is the name of the snapshot to create, apply or delete\n"
213            "  '-a' applies a snapshot (revert disk to saved state)\n"
214            "  '-c' creates a snapshot\n"
215            "  '-d' deletes a snapshot\n"
216            "  '-l' lists all snapshots in the given image\n"
217            "\n"
218            "Parameters to compare subcommand:\n"
219            "  '-f' first image format\n"
220            "  '-F' second image format\n"
221            "  '-s' run in Strict mode - fail on different image size or sector allocation\n"
222            "\n"
223            "Parameters to dd subcommand:\n"
224            "  'bs=BYTES' read and write up to BYTES bytes at a time "
225            "(default: 512)\n"
226            "  'count=N' copy only N input blocks\n"
227            "  'if=FILE' read from FILE\n"
228            "  'of=FILE' write to FILE\n"
229            "  'skip=N' skip N bs-sized blocks at the start of input\n";
230 
231     printf("%s\nSupported formats:", help_msg);
232     bdrv_iterate_format(format_print, NULL, false);
233     printf("\n\n" QEMU_HELP_BOTTOM "\n");
234     exit(EXIT_SUCCESS);
235 }
236 
237 /*
238  * Is @optarg safe for accumulate_options()?
239  * It is when multiple of them can be joined together separated by ','.
240  * To make that work, @optarg must not start with ',' (or else a
241  * separating ',' preceding it gets escaped), and it must not end with
242  * an odd number of ',' (or else a separating ',' following it gets
243  * escaped), or be empty (or else a separating ',' preceding it can
244  * escape a separating ',' following it).
245  *
246  */
247 static bool is_valid_option_list(const char *optarg)
248 {
249     size_t len = strlen(optarg);
250     size_t i;
251 
252     if (!optarg[0] || optarg[0] == ',') {
253         return false;
254     }
255 
256     for (i = len; i > 0 && optarg[i - 1] == ','; i--) {
257     }
258     if ((len - i) % 2) {
259         return false;
260     }
261 
262     return true;
263 }
264 
265 static int accumulate_options(char **options, char *optarg)
266 {
267     char *new_options;
268 
269     if (!is_valid_option_list(optarg)) {
270         error_report("Invalid option list: %s", optarg);
271         return -1;
272     }
273 
274     if (!*options) {
275         *options = g_strdup(optarg);
276     } else {
277         new_options = g_strdup_printf("%s,%s", *options, optarg);
278         g_free(*options);
279         *options = new_options;
280     }
281     return 0;
282 }
283 
284 static QemuOptsList qemu_source_opts = {
285     .name = "source",
286     .implied_opt_name = "file",
287     .head = QTAILQ_HEAD_INITIALIZER(qemu_source_opts.head),
288     .desc = {
289         { }
290     },
291 };
292 
293 static int G_GNUC_PRINTF(2, 3) qprintf(bool quiet, const char *fmt, ...)
294 {
295     int ret = 0;
296     if (!quiet) {
297         va_list args;
298         va_start(args, fmt);
299         ret = vprintf(fmt, args);
300         va_end(args);
301     }
302     return ret;
303 }
304 
305 
306 static int print_block_option_help(const char *filename, const char *fmt)
307 {
308     BlockDriver *drv, *proto_drv;
309     QemuOptsList *create_opts = NULL;
310     Error *local_err = NULL;
311 
312     /* Find driver and parse its options */
313     drv = bdrv_find_format(fmt);
314     if (!drv) {
315         error_report("Unknown file format '%s'", fmt);
316         return 1;
317     }
318 
319     if (!drv->create_opts) {
320         error_report("Format driver '%s' does not support image creation", fmt);
321         return 1;
322     }
323 
324     create_opts = qemu_opts_append(create_opts, drv->create_opts);
325     if (filename) {
326         proto_drv = bdrv_find_protocol(filename, true, &local_err);
327         if (!proto_drv) {
328             error_report_err(local_err);
329             qemu_opts_free(create_opts);
330             return 1;
331         }
332         if (!proto_drv->create_opts) {
333             error_report("Protocol driver '%s' does not support image creation",
334                          proto_drv->format_name);
335             qemu_opts_free(create_opts);
336             return 1;
337         }
338         create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
339     }
340 
341     if (filename) {
342         printf("Supported options:\n");
343     } else {
344         printf("Supported %s options:\n", fmt);
345     }
346     qemu_opts_print_help(create_opts, false);
347     qemu_opts_free(create_opts);
348 
349     if (!filename) {
350         printf("\n"
351                "The protocol level may support further options.\n"
352                "Specify the target filename to include those options.\n");
353     }
354 
355     return 0;
356 }
357 
358 
359 static BlockBackend *img_open_opts(const char *optstr,
360                                    QemuOpts *opts, int flags, bool writethrough,
361                                    bool quiet, bool force_share)
362 {
363     QDict *options;
364     Error *local_err = NULL;
365     BlockBackend *blk;
366     options = qemu_opts_to_qdict(opts, NULL);
367     if (force_share) {
368         if (qdict_haskey(options, BDRV_OPT_FORCE_SHARE)
369             && strcmp(qdict_get_str(options, BDRV_OPT_FORCE_SHARE), "on")) {
370             error_report("--force-share/-U conflicts with image options");
371             qobject_unref(options);
372             return NULL;
373         }
374         qdict_put_str(options, BDRV_OPT_FORCE_SHARE, "on");
375     }
376     blk = blk_new_open(NULL, NULL, options, flags, &local_err);
377     if (!blk) {
378         error_reportf_err(local_err, "Could not open '%s': ", optstr);
379         return NULL;
380     }
381     blk_set_enable_write_cache(blk, !writethrough);
382 
383     return blk;
384 }
385 
386 static BlockBackend *img_open_file(const char *filename,
387                                    QDict *options,
388                                    const char *fmt, int flags,
389                                    bool writethrough, bool quiet,
390                                    bool force_share)
391 {
392     BlockBackend *blk;
393     Error *local_err = NULL;
394 
395     if (!options) {
396         options = qdict_new();
397     }
398     if (fmt) {
399         qdict_put_str(options, "driver", fmt);
400     }
401 
402     if (force_share) {
403         qdict_put_bool(options, BDRV_OPT_FORCE_SHARE, true);
404     }
405     blk = blk_new_open(filename, NULL, options, flags, &local_err);
406     if (!blk) {
407         error_reportf_err(local_err, "Could not open '%s': ", filename);
408         return NULL;
409     }
410     blk_set_enable_write_cache(blk, !writethrough);
411 
412     return blk;
413 }
414 
415 
416 static int img_add_key_secrets(void *opaque,
417                                const char *name, const char *value,
418                                Error **errp)
419 {
420     QDict *options = opaque;
421 
422     if (g_str_has_suffix(name, "key-secret")) {
423         qdict_put_str(options, name, value);
424     }
425 
426     return 0;
427 }
428 
429 
430 static BlockBackend *img_open(bool image_opts,
431                               const char *filename,
432                               const char *fmt, int flags, bool writethrough,
433                               bool quiet, bool force_share)
434 {
435     BlockBackend *blk;
436     if (image_opts) {
437         QemuOpts *opts;
438         if (fmt) {
439             error_report("--image-opts and --format are mutually exclusive");
440             return NULL;
441         }
442         opts = qemu_opts_parse_noisily(qemu_find_opts("source"),
443                                        filename, true);
444         if (!opts) {
445             return NULL;
446         }
447         blk = img_open_opts(filename, opts, flags, writethrough, quiet,
448                             force_share);
449     } else {
450         blk = img_open_file(filename, NULL, fmt, flags, writethrough, quiet,
451                             force_share);
452     }
453     return blk;
454 }
455 
456 
457 static int add_old_style_options(const char *fmt, QemuOpts *opts,
458                                  const char *base_filename,
459                                  const char *base_fmt)
460 {
461     if (base_filename) {
462         if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
463                           NULL)) {
464             error_report("Backing file not supported for file format '%s'",
465                          fmt);
466             return -1;
467         }
468     }
469     if (base_fmt) {
470         if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
471             error_report("Backing file format not supported for file "
472                          "format '%s'", fmt);
473             return -1;
474         }
475     }
476     return 0;
477 }
478 
479 static int64_t cvtnum_full(const char *name, const char *value, int64_t min,
480                            int64_t max)
481 {
482     int err;
483     uint64_t res;
484 
485     err = qemu_strtosz(value, NULL, &res);
486     if (err < 0 && err != -ERANGE) {
487         error_report("Invalid %s specified. You may use "
488                      "k, M, G, T, P or E suffixes for", name);
489         error_report("kilobytes, megabytes, gigabytes, terabytes, "
490                      "petabytes and exabytes.");
491         return err;
492     }
493     if (err == -ERANGE || res > max || res < min) {
494         error_report("Invalid %s specified. Must be between %" PRId64
495                      " and %" PRId64 ".", name, min, max);
496         return -ERANGE;
497     }
498     return res;
499 }
500 
501 static int64_t cvtnum(const char *name, const char *value)
502 {
503     return cvtnum_full(name, value, 0, INT64_MAX);
504 }
505 
506 static int img_create(int argc, char **argv)
507 {
508     int c;
509     uint64_t img_size = -1;
510     const char *fmt = "raw";
511     const char *base_fmt = NULL;
512     const char *filename;
513     const char *base_filename = NULL;
514     char *options = NULL;
515     Error *local_err = NULL;
516     bool quiet = false;
517     int flags = 0;
518 
519     for(;;) {
520         static const struct option long_options[] = {
521             {"help", no_argument, 0, 'h'},
522             {"object", required_argument, 0, OPTION_OBJECT},
523             {0, 0, 0, 0}
524         };
525         c = getopt_long(argc, argv, ":F:b:f:ho:qu",
526                         long_options, NULL);
527         if (c == -1) {
528             break;
529         }
530         switch(c) {
531         case ':':
532             missing_argument(argv[optind - 1]);
533             break;
534         case '?':
535             unrecognized_option(argv[optind - 1]);
536             break;
537         case 'h':
538             help();
539             break;
540         case 'F':
541             base_fmt = optarg;
542             break;
543         case 'b':
544             base_filename = optarg;
545             break;
546         case 'f':
547             fmt = optarg;
548             break;
549         case 'o':
550             if (accumulate_options(&options, optarg) < 0) {
551                 goto fail;
552             }
553             break;
554         case 'q':
555             quiet = true;
556             break;
557         case 'u':
558             flags |= BDRV_O_NO_BACKING;
559             break;
560         case OPTION_OBJECT:
561             user_creatable_process_cmdline(optarg);
562             break;
563         }
564     }
565 
566     /* Get the filename */
567     filename = (optind < argc) ? argv[optind] : NULL;
568     if (options && has_help_option(options)) {
569         g_free(options);
570         return print_block_option_help(filename, fmt);
571     }
572 
573     if (optind >= argc) {
574         error_exit("Expecting image file name");
575     }
576     optind++;
577 
578     /* Get image size, if specified */
579     if (optind < argc) {
580         int64_t sval;
581 
582         sval = cvtnum("image size", argv[optind++]);
583         if (sval < 0) {
584             goto fail;
585         }
586         img_size = (uint64_t)sval;
587     }
588     if (optind != argc) {
589         error_exit("Unexpected argument: %s", argv[optind]);
590     }
591 
592     bdrv_img_create(filename, fmt, base_filename, base_fmt,
593                     options, img_size, flags, quiet, &local_err);
594     if (local_err) {
595         error_reportf_err(local_err, "%s: ", filename);
596         goto fail;
597     }
598 
599     g_free(options);
600     return 0;
601 
602 fail:
603     g_free(options);
604     return 1;
605 }
606 
607 static void dump_json_image_check(ImageCheck *check, bool quiet)
608 {
609     GString *str;
610     QObject *obj;
611     Visitor *v = qobject_output_visitor_new(&obj);
612 
613     visit_type_ImageCheck(v, NULL, &check, &error_abort);
614     visit_complete(v, &obj);
615     str = qobject_to_json_pretty(obj, true);
616     assert(str != NULL);
617     qprintf(quiet, "%s\n", str->str);
618     qobject_unref(obj);
619     visit_free(v);
620     g_string_free(str, true);
621 }
622 
623 static void dump_human_image_check(ImageCheck *check, bool quiet)
624 {
625     if (!(check->corruptions || check->leaks || check->check_errors)) {
626         qprintf(quiet, "No errors were found on the image.\n");
627     } else {
628         if (check->corruptions) {
629             qprintf(quiet, "\n%" PRId64 " errors were found on the image.\n"
630                     "Data may be corrupted, or further writes to the image "
631                     "may corrupt it.\n",
632                     check->corruptions);
633         }
634 
635         if (check->leaks) {
636             qprintf(quiet,
637                     "\n%" PRId64 " leaked clusters were found on the image.\n"
638                     "This means waste of disk space, but no harm to data.\n",
639                     check->leaks);
640         }
641 
642         if (check->check_errors) {
643             qprintf(quiet,
644                     "\n%" PRId64
645                     " internal errors have occurred during the check.\n",
646                     check->check_errors);
647         }
648     }
649 
650     if (check->total_clusters != 0 && check->allocated_clusters != 0) {
651         qprintf(quiet, "%" PRId64 "/%" PRId64 " = %0.2f%% allocated, "
652                 "%0.2f%% fragmented, %0.2f%% compressed clusters\n",
653                 check->allocated_clusters, check->total_clusters,
654                 check->allocated_clusters * 100.0 / check->total_clusters,
655                 check->fragmented_clusters * 100.0 / check->allocated_clusters,
656                 check->compressed_clusters * 100.0 /
657                 check->allocated_clusters);
658     }
659 
660     if (check->image_end_offset) {
661         qprintf(quiet,
662                 "Image end offset: %" PRId64 "\n", check->image_end_offset);
663     }
664 }
665 
666 static int collect_image_check(BlockDriverState *bs,
667                    ImageCheck *check,
668                    const char *filename,
669                    const char *fmt,
670                    int fix)
671 {
672     int ret;
673     BdrvCheckResult result;
674 
675     ret = bdrv_check(bs, &result, fix);
676     if (ret < 0) {
677         return ret;
678     }
679 
680     check->filename                 = g_strdup(filename);
681     check->format                   = g_strdup(bdrv_get_format_name(bs));
682     check->check_errors             = result.check_errors;
683     check->corruptions              = result.corruptions;
684     check->has_corruptions          = result.corruptions != 0;
685     check->leaks                    = result.leaks;
686     check->has_leaks                = result.leaks != 0;
687     check->corruptions_fixed        = result.corruptions_fixed;
688     check->has_corruptions_fixed    = result.corruptions_fixed != 0;
689     check->leaks_fixed              = result.leaks_fixed;
690     check->has_leaks_fixed          = result.leaks_fixed != 0;
691     check->image_end_offset         = result.image_end_offset;
692     check->has_image_end_offset     = result.image_end_offset != 0;
693     check->total_clusters           = result.bfi.total_clusters;
694     check->has_total_clusters       = result.bfi.total_clusters != 0;
695     check->allocated_clusters       = result.bfi.allocated_clusters;
696     check->has_allocated_clusters   = result.bfi.allocated_clusters != 0;
697     check->fragmented_clusters      = result.bfi.fragmented_clusters;
698     check->has_fragmented_clusters  = result.bfi.fragmented_clusters != 0;
699     check->compressed_clusters      = result.bfi.compressed_clusters;
700     check->has_compressed_clusters  = result.bfi.compressed_clusters != 0;
701 
702     return 0;
703 }
704 
705 /*
706  * Checks an image for consistency. Exit codes:
707  *
708  *  0 - Check completed, image is good
709  *  1 - Check not completed because of internal errors
710  *  2 - Check completed, image is corrupted
711  *  3 - Check completed, image has leaked clusters, but is good otherwise
712  * 63 - Checks are not supported by the image format
713  */
714 static int img_check(int argc, char **argv)
715 {
716     int c, ret;
717     OutputFormat output_format = OFORMAT_HUMAN;
718     const char *filename, *fmt, *output, *cache;
719     BlockBackend *blk;
720     BlockDriverState *bs;
721     int fix = 0;
722     int flags = BDRV_O_CHECK;
723     bool writethrough;
724     ImageCheck *check;
725     bool quiet = false;
726     bool image_opts = false;
727     bool force_share = false;
728 
729     fmt = NULL;
730     output = NULL;
731     cache = BDRV_DEFAULT_CACHE;
732 
733     for(;;) {
734         int option_index = 0;
735         static const struct option long_options[] = {
736             {"help", no_argument, 0, 'h'},
737             {"format", required_argument, 0, 'f'},
738             {"repair", required_argument, 0, 'r'},
739             {"output", required_argument, 0, OPTION_OUTPUT},
740             {"object", required_argument, 0, OPTION_OBJECT},
741             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
742             {"force-share", no_argument, 0, 'U'},
743             {0, 0, 0, 0}
744         };
745         c = getopt_long(argc, argv, ":hf:r:T:qU",
746                         long_options, &option_index);
747         if (c == -1) {
748             break;
749         }
750         switch(c) {
751         case ':':
752             missing_argument(argv[optind - 1]);
753             break;
754         case '?':
755             unrecognized_option(argv[optind - 1]);
756             break;
757         case 'h':
758             help();
759             break;
760         case 'f':
761             fmt = optarg;
762             break;
763         case 'r':
764             flags |= BDRV_O_RDWR;
765 
766             if (!strcmp(optarg, "leaks")) {
767                 fix = BDRV_FIX_LEAKS;
768             } else if (!strcmp(optarg, "all")) {
769                 fix = BDRV_FIX_LEAKS | BDRV_FIX_ERRORS;
770             } else {
771                 error_exit("Unknown option value for -r "
772                            "(expecting 'leaks' or 'all'): %s", optarg);
773             }
774             break;
775         case OPTION_OUTPUT:
776             output = optarg;
777             break;
778         case 'T':
779             cache = optarg;
780             break;
781         case 'q':
782             quiet = true;
783             break;
784         case 'U':
785             force_share = true;
786             break;
787         case OPTION_OBJECT:
788             user_creatable_process_cmdline(optarg);
789             break;
790         case OPTION_IMAGE_OPTS:
791             image_opts = true;
792             break;
793         }
794     }
795     if (optind != argc - 1) {
796         error_exit("Expecting one image file name");
797     }
798     filename = argv[optind++];
799 
800     if (output && !strcmp(output, "json")) {
801         output_format = OFORMAT_JSON;
802     } else if (output && !strcmp(output, "human")) {
803         output_format = OFORMAT_HUMAN;
804     } else if (output) {
805         error_report("--output must be used with human or json as argument.");
806         return 1;
807     }
808 
809     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
810     if (ret < 0) {
811         error_report("Invalid source cache option: %s", cache);
812         return 1;
813     }
814 
815     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
816                    force_share);
817     if (!blk) {
818         return 1;
819     }
820     bs = blk_bs(blk);
821 
822     check = g_new0(ImageCheck, 1);
823     ret = collect_image_check(bs, check, filename, fmt, fix);
824 
825     if (ret == -ENOTSUP) {
826         error_report("This image format does not support checks");
827         ret = 63;
828         goto fail;
829     }
830 
831     if (check->corruptions_fixed || check->leaks_fixed) {
832         int corruptions_fixed, leaks_fixed;
833         bool has_leaks_fixed, has_corruptions_fixed;
834 
835         leaks_fixed         = check->leaks_fixed;
836         has_leaks_fixed     = check->has_leaks_fixed;
837         corruptions_fixed   = check->corruptions_fixed;
838         has_corruptions_fixed = check->has_corruptions_fixed;
839 
840         if (output_format == OFORMAT_HUMAN) {
841             qprintf(quiet,
842                     "The following inconsistencies were found and repaired:\n\n"
843                     "    %" PRId64 " leaked clusters\n"
844                     "    %" PRId64 " corruptions\n\n"
845                     "Double checking the fixed image now...\n",
846                     check->leaks_fixed,
847                     check->corruptions_fixed);
848         }
849 
850         qapi_free_ImageCheck(check);
851         check = g_new0(ImageCheck, 1);
852         ret = collect_image_check(bs, check, filename, fmt, 0);
853 
854         check->leaks_fixed          = leaks_fixed;
855         check->has_leaks_fixed      = has_leaks_fixed;
856         check->corruptions_fixed    = corruptions_fixed;
857         check->has_corruptions_fixed = has_corruptions_fixed;
858     }
859 
860     if (!ret) {
861         switch (output_format) {
862         case OFORMAT_HUMAN:
863             dump_human_image_check(check, quiet);
864             break;
865         case OFORMAT_JSON:
866             dump_json_image_check(check, quiet);
867             break;
868         }
869     }
870 
871     if (ret || check->check_errors) {
872         if (ret) {
873             error_report("Check failed: %s", strerror(-ret));
874         } else {
875             error_report("Check failed");
876         }
877         ret = 1;
878         goto fail;
879     }
880 
881     if (check->corruptions) {
882         ret = 2;
883     } else if (check->leaks) {
884         ret = 3;
885     } else {
886         ret = 0;
887     }
888 
889 fail:
890     qapi_free_ImageCheck(check);
891     blk_unref(blk);
892     return ret;
893 }
894 
895 typedef struct CommonBlockJobCBInfo {
896     BlockDriverState *bs;
897     Error **errp;
898 } CommonBlockJobCBInfo;
899 
900 static void common_block_job_cb(void *opaque, int ret)
901 {
902     CommonBlockJobCBInfo *cbi = opaque;
903 
904     if (ret < 0) {
905         error_setg_errno(cbi->errp, -ret, "Block job failed");
906     }
907 }
908 
909 static void run_block_job(BlockJob *job, Error **errp)
910 {
911     uint64_t progress_current, progress_total;
912     AioContext *aio_context = block_job_get_aio_context(job);
913     int ret = 0;
914 
915     job_lock();
916     job_ref_locked(&job->job);
917     do {
918         float progress = 0.0f;
919         job_unlock();
920         aio_poll(aio_context, true);
921 
922         progress_get_snapshot(&job->job.progress, &progress_current,
923                               &progress_total);
924         if (progress_total) {
925             progress = (float)progress_current / progress_total * 100.f;
926         }
927         qemu_progress_print(progress, 0);
928         job_lock();
929     } while (!job_is_ready_locked(&job->job) &&
930              !job_is_completed_locked(&job->job));
931 
932     if (!job_is_completed_locked(&job->job)) {
933         ret = job_complete_sync_locked(&job->job, errp);
934     } else {
935         ret = job->job.ret;
936     }
937     job_unref_locked(&job->job);
938     job_unlock();
939 
940     /* publish completion progress only when success */
941     if (!ret) {
942         qemu_progress_print(100.f, 0);
943     }
944 }
945 
946 static int img_commit(int argc, char **argv)
947 {
948     int c, ret, flags;
949     const char *filename, *fmt, *cache, *base;
950     BlockBackend *blk;
951     BlockDriverState *bs, *base_bs;
952     BlockJob *job;
953     bool progress = false, quiet = false, drop = false;
954     bool writethrough;
955     Error *local_err = NULL;
956     CommonBlockJobCBInfo cbi;
957     bool image_opts = false;
958     AioContext *aio_context;
959     int64_t rate_limit = 0;
960 
961     fmt = NULL;
962     cache = BDRV_DEFAULT_CACHE;
963     base = NULL;
964     for(;;) {
965         static const struct option long_options[] = {
966             {"help", no_argument, 0, 'h'},
967             {"object", required_argument, 0, OPTION_OBJECT},
968             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
969             {0, 0, 0, 0}
970         };
971         c = getopt_long(argc, argv, ":f:ht:b:dpqr:",
972                         long_options, NULL);
973         if (c == -1) {
974             break;
975         }
976         switch(c) {
977         case ':':
978             missing_argument(argv[optind - 1]);
979             break;
980         case '?':
981             unrecognized_option(argv[optind - 1]);
982             break;
983         case 'h':
984             help();
985             break;
986         case 'f':
987             fmt = optarg;
988             break;
989         case 't':
990             cache = optarg;
991             break;
992         case 'b':
993             base = optarg;
994             /* -b implies -d */
995             drop = true;
996             break;
997         case 'd':
998             drop = true;
999             break;
1000         case 'p':
1001             progress = true;
1002             break;
1003         case 'q':
1004             quiet = true;
1005             break;
1006         case 'r':
1007             rate_limit = cvtnum("rate limit", optarg);
1008             if (rate_limit < 0) {
1009                 return 1;
1010             }
1011             break;
1012         case OPTION_OBJECT:
1013             user_creatable_process_cmdline(optarg);
1014             break;
1015         case OPTION_IMAGE_OPTS:
1016             image_opts = true;
1017             break;
1018         }
1019     }
1020 
1021     /* Progress is not shown in Quiet mode */
1022     if (quiet) {
1023         progress = false;
1024     }
1025 
1026     if (optind != argc - 1) {
1027         error_exit("Expecting one image file name");
1028     }
1029     filename = argv[optind++];
1030 
1031     flags = BDRV_O_RDWR | BDRV_O_UNMAP;
1032     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
1033     if (ret < 0) {
1034         error_report("Invalid cache option: %s", cache);
1035         return 1;
1036     }
1037 
1038     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
1039                    false);
1040     if (!blk) {
1041         return 1;
1042     }
1043     bs = blk_bs(blk);
1044 
1045     qemu_progress_init(progress, 1.f);
1046     qemu_progress_print(0.f, 100);
1047 
1048     if (base) {
1049         base_bs = bdrv_find_backing_image(bs, base);
1050         if (!base_bs) {
1051             error_setg(&local_err,
1052                        "Did not find '%s' in the backing chain of '%s'",
1053                        base, filename);
1054             goto done;
1055         }
1056     } else {
1057         /* This is different from QMP, which by default uses the deepest file in
1058          * the backing chain (i.e., the very base); however, the traditional
1059          * behavior of qemu-img commit is using the immediate backing file. */
1060         base_bs = bdrv_backing_chain_next(bs);
1061         if (!base_bs) {
1062             error_setg(&local_err, "Image does not have a backing file");
1063             goto done;
1064         }
1065     }
1066 
1067     cbi = (CommonBlockJobCBInfo){
1068         .errp = &local_err,
1069         .bs   = bs,
1070     };
1071 
1072     aio_context = bdrv_get_aio_context(bs);
1073     aio_context_acquire(aio_context);
1074     commit_active_start("commit", bs, base_bs, JOB_DEFAULT, rate_limit,
1075                         BLOCKDEV_ON_ERROR_REPORT, NULL, common_block_job_cb,
1076                         &cbi, false, &local_err);
1077     aio_context_release(aio_context);
1078     if (local_err) {
1079         goto done;
1080     }
1081 
1082     /* When the block job completes, the BlockBackend reference will point to
1083      * the old backing file. In order to avoid that the top image is already
1084      * deleted, so we can still empty it afterwards, increment the reference
1085      * counter here preemptively. */
1086     if (!drop) {
1087         bdrv_ref(bs);
1088     }
1089 
1090     job = block_job_get("commit");
1091     assert(job);
1092     run_block_job(job, &local_err);
1093     if (local_err) {
1094         goto unref_backing;
1095     }
1096 
1097     if (!drop) {
1098         BlockBackend *old_backing_blk;
1099 
1100         old_backing_blk = blk_new_with_bs(bs, BLK_PERM_WRITE, BLK_PERM_ALL,
1101                                           &local_err);
1102         if (!old_backing_blk) {
1103             goto unref_backing;
1104         }
1105         ret = blk_make_empty(old_backing_blk, &local_err);
1106         blk_unref(old_backing_blk);
1107         if (ret == -ENOTSUP) {
1108             error_free(local_err);
1109             local_err = NULL;
1110         } else if (ret < 0) {
1111             goto unref_backing;
1112         }
1113     }
1114 
1115 unref_backing:
1116     if (!drop) {
1117         bdrv_unref(bs);
1118     }
1119 
1120 done:
1121     qemu_progress_end();
1122 
1123     blk_unref(blk);
1124 
1125     if (local_err) {
1126         error_report_err(local_err);
1127         return 1;
1128     }
1129 
1130     qprintf(quiet, "Image committed.\n");
1131     return 0;
1132 }
1133 
1134 /*
1135  * Returns -1 if 'buf' contains only zeroes, otherwise the byte index
1136  * of the first sector boundary within buf where the sector contains a
1137  * non-zero byte.  This function is robust to a buffer that is not
1138  * sector-aligned.
1139  */
1140 static int64_t find_nonzero(const uint8_t *buf, int64_t n)
1141 {
1142     int64_t i;
1143     int64_t end = QEMU_ALIGN_DOWN(n, BDRV_SECTOR_SIZE);
1144 
1145     for (i = 0; i < end; i += BDRV_SECTOR_SIZE) {
1146         if (!buffer_is_zero(buf + i, BDRV_SECTOR_SIZE)) {
1147             return i;
1148         }
1149     }
1150     if (i < n && !buffer_is_zero(buf + i, n - end)) {
1151         return i;
1152     }
1153     return -1;
1154 }
1155 
1156 /*
1157  * Returns true iff the first sector pointed to by 'buf' contains at least
1158  * a non-NUL byte.
1159  *
1160  * 'pnum' is set to the number of sectors (including and immediately following
1161  * the first one) that are known to be in the same allocated/unallocated state.
1162  * The function will try to align the end offset to alignment boundaries so
1163  * that the request will at least end aligned and consecutive requests will
1164  * also start at an aligned offset.
1165  */
1166 static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum,
1167                                 int64_t sector_num, int alignment)
1168 {
1169     bool is_zero;
1170     int i, tail;
1171 
1172     if (n <= 0) {
1173         *pnum = 0;
1174         return 0;
1175     }
1176     is_zero = buffer_is_zero(buf, BDRV_SECTOR_SIZE);
1177     for(i = 1; i < n; i++) {
1178         buf += BDRV_SECTOR_SIZE;
1179         if (is_zero != buffer_is_zero(buf, BDRV_SECTOR_SIZE)) {
1180             break;
1181         }
1182     }
1183 
1184     if (i == n) {
1185         /*
1186          * The whole buf is the same.
1187          * No reason to split it into chunks, so return now.
1188          */
1189         *pnum = i;
1190         return !is_zero;
1191     }
1192 
1193     tail = (sector_num + i) & (alignment - 1);
1194     if (tail) {
1195         if (is_zero && i <= tail) {
1196             /*
1197              * For sure next sector after i is data, and it will rewrite this
1198              * tail anyway due to RMW. So, let's just write data now.
1199              */
1200             is_zero = false;
1201         }
1202         if (!is_zero) {
1203             /* If possible, align up end offset of allocated areas. */
1204             i += alignment - tail;
1205             i = MIN(i, n);
1206         } else {
1207             /*
1208              * For sure next sector after i is data, and it will rewrite this
1209              * tail anyway due to RMW. Better is avoid RMW and write zeroes up
1210              * to aligned bound.
1211              */
1212             i -= tail;
1213         }
1214     }
1215     *pnum = i;
1216     return !is_zero;
1217 }
1218 
1219 /*
1220  * Like is_allocated_sectors, but if the buffer starts with a used sector,
1221  * up to 'min' consecutive sectors containing zeros are ignored. This avoids
1222  * breaking up write requests for only small sparse areas.
1223  */
1224 static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum,
1225     int min, int64_t sector_num, int alignment)
1226 {
1227     int ret;
1228     int num_checked, num_used;
1229 
1230     if (n < min) {
1231         min = n;
1232     }
1233 
1234     ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
1235     if (!ret) {
1236         return ret;
1237     }
1238 
1239     num_used = *pnum;
1240     buf += BDRV_SECTOR_SIZE * *pnum;
1241     n -= *pnum;
1242     sector_num += *pnum;
1243     num_checked = num_used;
1244 
1245     while (n > 0) {
1246         ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
1247 
1248         buf += BDRV_SECTOR_SIZE * *pnum;
1249         n -= *pnum;
1250         sector_num += *pnum;
1251         num_checked += *pnum;
1252         if (ret) {
1253             num_used = num_checked;
1254         } else if (*pnum >= min) {
1255             break;
1256         }
1257     }
1258 
1259     *pnum = num_used;
1260     return 1;
1261 }
1262 
1263 /*
1264  * Compares two buffers sector by sector. Returns 0 if the first
1265  * sector of each buffer matches, non-zero otherwise.
1266  *
1267  * pnum is set to the sector-aligned size of the buffer prefix that
1268  * has the same matching status as the first sector.
1269  */
1270 static int compare_buffers(const uint8_t *buf1, const uint8_t *buf2,
1271                            int64_t bytes, int64_t *pnum)
1272 {
1273     bool res;
1274     int64_t i = MIN(bytes, BDRV_SECTOR_SIZE);
1275 
1276     assert(bytes > 0);
1277 
1278     res = !!memcmp(buf1, buf2, i);
1279     while (i < bytes) {
1280         int64_t len = MIN(bytes - i, BDRV_SECTOR_SIZE);
1281 
1282         if (!!memcmp(buf1 + i, buf2 + i, len) != res) {
1283             break;
1284         }
1285         i += len;
1286     }
1287 
1288     *pnum = i;
1289     return res;
1290 }
1291 
1292 #define IO_BUF_SIZE (2 * MiB)
1293 
1294 /*
1295  * Check if passed sectors are empty (not allocated or contain only 0 bytes)
1296  *
1297  * Intended for use by 'qemu-img compare': Returns 0 in case sectors are
1298  * filled with 0, 1 if sectors contain non-zero data (this is a comparison
1299  * failure), and 4 on error (the exit status for read errors), after emitting
1300  * an error message.
1301  *
1302  * @param blk:  BlockBackend for the image
1303  * @param offset: Starting offset to check
1304  * @param bytes: Number of bytes to check
1305  * @param filename: Name of disk file we are checking (logging purpose)
1306  * @param buffer: Allocated buffer for storing read data
1307  * @param quiet: Flag for quiet mode
1308  */
1309 static int check_empty_sectors(BlockBackend *blk, int64_t offset,
1310                                int64_t bytes, const char *filename,
1311                                uint8_t *buffer, bool quiet)
1312 {
1313     int ret = 0;
1314     int64_t idx;
1315 
1316     ret = blk_pread(blk, offset, bytes, buffer, 0);
1317     if (ret < 0) {
1318         error_report("Error while reading offset %" PRId64 " of %s: %s",
1319                      offset, filename, strerror(-ret));
1320         return 4;
1321     }
1322     idx = find_nonzero(buffer, bytes);
1323     if (idx >= 0) {
1324         qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1325                 offset + idx);
1326         return 1;
1327     }
1328 
1329     return 0;
1330 }
1331 
1332 /*
1333  * Compares two images. Exit codes:
1334  *
1335  * 0 - Images are identical or the requested help was printed
1336  * 1 - Images differ
1337  * >1 - Error occurred
1338  */
1339 static int img_compare(int argc, char **argv)
1340 {
1341     const char *fmt1 = NULL, *fmt2 = NULL, *cache, *filename1, *filename2;
1342     BlockBackend *blk1, *blk2;
1343     BlockDriverState *bs1, *bs2;
1344     int64_t total_size1, total_size2;
1345     uint8_t *buf1 = NULL, *buf2 = NULL;
1346     int64_t pnum1, pnum2;
1347     int allocated1, allocated2;
1348     int ret = 0; /* return value - 0 Ident, 1 Different, >1 Error */
1349     bool progress = false, quiet = false, strict = false;
1350     int flags;
1351     bool writethrough;
1352     int64_t total_size;
1353     int64_t offset = 0;
1354     int64_t chunk;
1355     int c;
1356     uint64_t progress_base;
1357     bool image_opts = false;
1358     bool force_share = false;
1359 
1360     cache = BDRV_DEFAULT_CACHE;
1361     for (;;) {
1362         static const struct option long_options[] = {
1363             {"help", no_argument, 0, 'h'},
1364             {"object", required_argument, 0, OPTION_OBJECT},
1365             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
1366             {"force-share", no_argument, 0, 'U'},
1367             {0, 0, 0, 0}
1368         };
1369         c = getopt_long(argc, argv, ":hf:F:T:pqsU",
1370                         long_options, NULL);
1371         if (c == -1) {
1372             break;
1373         }
1374         switch (c) {
1375         case ':':
1376             missing_argument(argv[optind - 1]);
1377             break;
1378         case '?':
1379             unrecognized_option(argv[optind - 1]);
1380             break;
1381         case 'h':
1382             help();
1383             break;
1384         case 'f':
1385             fmt1 = optarg;
1386             break;
1387         case 'F':
1388             fmt2 = optarg;
1389             break;
1390         case 'T':
1391             cache = optarg;
1392             break;
1393         case 'p':
1394             progress = true;
1395             break;
1396         case 'q':
1397             quiet = true;
1398             break;
1399         case 's':
1400             strict = true;
1401             break;
1402         case 'U':
1403             force_share = true;
1404             break;
1405         case OPTION_OBJECT:
1406             {
1407                 Error *local_err = NULL;
1408 
1409                 if (!user_creatable_add_from_str(optarg, &local_err)) {
1410                     if (local_err) {
1411                         error_report_err(local_err);
1412                         exit(2);
1413                     } else {
1414                         /* Help was printed */
1415                         exit(EXIT_SUCCESS);
1416                     }
1417                 }
1418                 break;
1419             }
1420         case OPTION_IMAGE_OPTS:
1421             image_opts = true;
1422             break;
1423         }
1424     }
1425 
1426     /* Progress is not shown in Quiet mode */
1427     if (quiet) {
1428         progress = false;
1429     }
1430 
1431 
1432     if (optind != argc - 2) {
1433         error_exit("Expecting two image file names");
1434     }
1435     filename1 = argv[optind++];
1436     filename2 = argv[optind++];
1437 
1438     /* Initialize before goto out */
1439     qemu_progress_init(progress, 2.0);
1440 
1441     flags = 0;
1442     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
1443     if (ret < 0) {
1444         error_report("Invalid source cache option: %s", cache);
1445         ret = 2;
1446         goto out3;
1447     }
1448 
1449     blk1 = img_open(image_opts, filename1, fmt1, flags, writethrough, quiet,
1450                     force_share);
1451     if (!blk1) {
1452         ret = 2;
1453         goto out3;
1454     }
1455 
1456     blk2 = img_open(image_opts, filename2, fmt2, flags, writethrough, quiet,
1457                     force_share);
1458     if (!blk2) {
1459         ret = 2;
1460         goto out2;
1461     }
1462     bs1 = blk_bs(blk1);
1463     bs2 = blk_bs(blk2);
1464 
1465     buf1 = blk_blockalign(blk1, IO_BUF_SIZE);
1466     buf2 = blk_blockalign(blk2, IO_BUF_SIZE);
1467     total_size1 = blk_getlength(blk1);
1468     if (total_size1 < 0) {
1469         error_report("Can't get size of %s: %s",
1470                      filename1, strerror(-total_size1));
1471         ret = 4;
1472         goto out;
1473     }
1474     total_size2 = blk_getlength(blk2);
1475     if (total_size2 < 0) {
1476         error_report("Can't get size of %s: %s",
1477                      filename2, strerror(-total_size2));
1478         ret = 4;
1479         goto out;
1480     }
1481     total_size = MIN(total_size1, total_size2);
1482     progress_base = MAX(total_size1, total_size2);
1483 
1484     qemu_progress_print(0, 100);
1485 
1486     if (strict && total_size1 != total_size2) {
1487         ret = 1;
1488         qprintf(quiet, "Strict mode: Image size mismatch!\n");
1489         goto out;
1490     }
1491 
1492     while (offset < total_size) {
1493         int status1, status2;
1494 
1495         status1 = bdrv_block_status_above(bs1, NULL, offset,
1496                                           total_size1 - offset, &pnum1, NULL,
1497                                           NULL);
1498         if (status1 < 0) {
1499             ret = 3;
1500             error_report("Sector allocation test failed for %s", filename1);
1501             goto out;
1502         }
1503         allocated1 = status1 & BDRV_BLOCK_ALLOCATED;
1504 
1505         status2 = bdrv_block_status_above(bs2, NULL, offset,
1506                                           total_size2 - offset, &pnum2, NULL,
1507                                           NULL);
1508         if (status2 < 0) {
1509             ret = 3;
1510             error_report("Sector allocation test failed for %s", filename2);
1511             goto out;
1512         }
1513         allocated2 = status2 & BDRV_BLOCK_ALLOCATED;
1514 
1515         assert(pnum1 && pnum2);
1516         chunk = MIN(pnum1, pnum2);
1517 
1518         if (strict) {
1519             if (status1 != status2) {
1520                 ret = 1;
1521                 qprintf(quiet, "Strict mode: Offset %" PRId64
1522                         " block status mismatch!\n", offset);
1523                 goto out;
1524             }
1525         }
1526         if ((status1 & BDRV_BLOCK_ZERO) && (status2 & BDRV_BLOCK_ZERO)) {
1527             /* nothing to do */
1528         } else if (allocated1 == allocated2) {
1529             if (allocated1) {
1530                 int64_t pnum;
1531 
1532                 chunk = MIN(chunk, IO_BUF_SIZE);
1533                 ret = blk_pread(blk1, offset, chunk, buf1, 0);
1534                 if (ret < 0) {
1535                     error_report("Error while reading offset %" PRId64
1536                                  " of %s: %s",
1537                                  offset, filename1, strerror(-ret));
1538                     ret = 4;
1539                     goto out;
1540                 }
1541                 ret = blk_pread(blk2, offset, chunk, buf2, 0);
1542                 if (ret < 0) {
1543                     error_report("Error while reading offset %" PRId64
1544                                  " of %s: %s",
1545                                  offset, filename2, strerror(-ret));
1546                     ret = 4;
1547                     goto out;
1548                 }
1549                 ret = compare_buffers(buf1, buf2, chunk, &pnum);
1550                 if (ret || pnum != chunk) {
1551                     qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1552                             offset + (ret ? 0 : pnum));
1553                     ret = 1;
1554                     goto out;
1555                 }
1556             }
1557         } else {
1558             chunk = MIN(chunk, IO_BUF_SIZE);
1559             if (allocated1) {
1560                 ret = check_empty_sectors(blk1, offset, chunk,
1561                                           filename1, buf1, quiet);
1562             } else {
1563                 ret = check_empty_sectors(blk2, offset, chunk,
1564                                           filename2, buf1, quiet);
1565             }
1566             if (ret) {
1567                 goto out;
1568             }
1569         }
1570         offset += chunk;
1571         qemu_progress_print(((float) chunk / progress_base) * 100, 100);
1572     }
1573 
1574     if (total_size1 != total_size2) {
1575         BlockBackend *blk_over;
1576         const char *filename_over;
1577 
1578         qprintf(quiet, "Warning: Image size mismatch!\n");
1579         if (total_size1 > total_size2) {
1580             blk_over = blk1;
1581             filename_over = filename1;
1582         } else {
1583             blk_over = blk2;
1584             filename_over = filename2;
1585         }
1586 
1587         while (offset < progress_base) {
1588             ret = bdrv_block_status_above(blk_bs(blk_over), NULL, offset,
1589                                           progress_base - offset, &chunk,
1590                                           NULL, NULL);
1591             if (ret < 0) {
1592                 ret = 3;
1593                 error_report("Sector allocation test failed for %s",
1594                              filename_over);
1595                 goto out;
1596 
1597             }
1598             if (ret & BDRV_BLOCK_ALLOCATED && !(ret & BDRV_BLOCK_ZERO)) {
1599                 chunk = MIN(chunk, IO_BUF_SIZE);
1600                 ret = check_empty_sectors(blk_over, offset, chunk,
1601                                           filename_over, buf1, quiet);
1602                 if (ret) {
1603                     goto out;
1604                 }
1605             }
1606             offset += chunk;
1607             qemu_progress_print(((float) chunk / progress_base) * 100, 100);
1608         }
1609     }
1610 
1611     qprintf(quiet, "Images are identical.\n");
1612     ret = 0;
1613 
1614 out:
1615     qemu_vfree(buf1);
1616     qemu_vfree(buf2);
1617     blk_unref(blk2);
1618 out2:
1619     blk_unref(blk1);
1620 out3:
1621     qemu_progress_end();
1622     return ret;
1623 }
1624 
1625 /* Convenience wrapper around qmp_block_dirty_bitmap_merge */
1626 static void do_dirty_bitmap_merge(const char *dst_node, const char *dst_name,
1627                                   const char *src_node, const char *src_name,
1628                                   Error **errp)
1629 {
1630     BlockDirtyBitmapOrStr *merge_src;
1631     BlockDirtyBitmapOrStrList *list = NULL;
1632 
1633     merge_src = g_new0(BlockDirtyBitmapOrStr, 1);
1634     merge_src->type = QTYPE_QDICT;
1635     merge_src->u.external.node = g_strdup(src_node);
1636     merge_src->u.external.name = g_strdup(src_name);
1637     QAPI_LIST_PREPEND(list, merge_src);
1638     qmp_block_dirty_bitmap_merge(dst_node, dst_name, list, errp);
1639     qapi_free_BlockDirtyBitmapOrStrList(list);
1640 }
1641 
1642 enum ImgConvertBlockStatus {
1643     BLK_DATA,
1644     BLK_ZERO,
1645     BLK_BACKING_FILE,
1646 };
1647 
1648 #define MAX_COROUTINES 16
1649 #define CONVERT_THROTTLE_GROUP "img_convert"
1650 
1651 typedef struct ImgConvertState {
1652     BlockBackend **src;
1653     int64_t *src_sectors;
1654     int *src_alignment;
1655     int src_num;
1656     int64_t total_sectors;
1657     int64_t allocated_sectors;
1658     int64_t allocated_done;
1659     int64_t sector_num;
1660     int64_t wr_offs;
1661     enum ImgConvertBlockStatus status;
1662     int64_t sector_next_status;
1663     BlockBackend *target;
1664     bool has_zero_init;
1665     bool compressed;
1666     bool target_is_new;
1667     bool target_has_backing;
1668     int64_t target_backing_sectors; /* negative if unknown */
1669     bool wr_in_order;
1670     bool copy_range;
1671     bool salvage;
1672     bool quiet;
1673     int min_sparse;
1674     int alignment;
1675     size_t cluster_sectors;
1676     size_t buf_sectors;
1677     long num_coroutines;
1678     int running_coroutines;
1679     Coroutine *co[MAX_COROUTINES];
1680     int64_t wait_sector_num[MAX_COROUTINES];
1681     CoMutex lock;
1682     int ret;
1683 } ImgConvertState;
1684 
1685 static void convert_select_part(ImgConvertState *s, int64_t sector_num,
1686                                 int *src_cur, int64_t *src_cur_offset)
1687 {
1688     *src_cur = 0;
1689     *src_cur_offset = 0;
1690     while (sector_num - *src_cur_offset >= s->src_sectors[*src_cur]) {
1691         *src_cur_offset += s->src_sectors[*src_cur];
1692         (*src_cur)++;
1693         assert(*src_cur < s->src_num);
1694     }
1695 }
1696 
1697 static int convert_iteration_sectors(ImgConvertState *s, int64_t sector_num)
1698 {
1699     int64_t src_cur_offset;
1700     int ret, n, src_cur;
1701     bool post_backing_zero = false;
1702 
1703     convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1704 
1705     assert(s->total_sectors > sector_num);
1706     n = MIN(s->total_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
1707 
1708     if (s->target_backing_sectors >= 0) {
1709         if (sector_num >= s->target_backing_sectors) {
1710             post_backing_zero = true;
1711         } else if (sector_num + n > s->target_backing_sectors) {
1712             /* Split requests around target_backing_sectors (because
1713              * starting from there, zeros are handled differently) */
1714             n = s->target_backing_sectors - sector_num;
1715         }
1716     }
1717 
1718     if (s->sector_next_status <= sector_num) {
1719         uint64_t offset = (sector_num - src_cur_offset) * BDRV_SECTOR_SIZE;
1720         int64_t count;
1721         int tail;
1722         BlockDriverState *src_bs = blk_bs(s->src[src_cur]);
1723         BlockDriverState *base;
1724 
1725         if (s->target_has_backing) {
1726             base = bdrv_cow_bs(bdrv_skip_filters(src_bs));
1727         } else {
1728             base = NULL;
1729         }
1730 
1731         do {
1732             count = n * BDRV_SECTOR_SIZE;
1733 
1734             ret = bdrv_block_status_above(src_bs, base, offset, count, &count,
1735                                           NULL, NULL);
1736 
1737             if (ret < 0) {
1738                 if (s->salvage) {
1739                     if (n == 1) {
1740                         if (!s->quiet) {
1741                             warn_report("error while reading block status at "
1742                                         "offset %" PRIu64 ": %s", offset,
1743                                         strerror(-ret));
1744                         }
1745                         /* Just try to read the data, then */
1746                         ret = BDRV_BLOCK_DATA;
1747                         count = BDRV_SECTOR_SIZE;
1748                     } else {
1749                         /* Retry on a shorter range */
1750                         n = DIV_ROUND_UP(n, 4);
1751                     }
1752                 } else {
1753                     error_report("error while reading block status at offset "
1754                                  "%" PRIu64 ": %s", offset, strerror(-ret));
1755                     return ret;
1756                 }
1757             }
1758         } while (ret < 0);
1759 
1760         n = DIV_ROUND_UP(count, BDRV_SECTOR_SIZE);
1761 
1762         /*
1763          * Avoid that s->sector_next_status becomes unaligned to the source
1764          * request alignment and/or cluster size to avoid unnecessary read
1765          * cycles.
1766          */
1767         tail = (sector_num - src_cur_offset + n) % s->src_alignment[src_cur];
1768         if (n > tail) {
1769             n -= tail;
1770         }
1771 
1772         if (ret & BDRV_BLOCK_ZERO) {
1773             s->status = post_backing_zero ? BLK_BACKING_FILE : BLK_ZERO;
1774         } else if (ret & BDRV_BLOCK_DATA) {
1775             s->status = BLK_DATA;
1776         } else {
1777             s->status = s->target_has_backing ? BLK_BACKING_FILE : BLK_DATA;
1778         }
1779 
1780         s->sector_next_status = sector_num + n;
1781     }
1782 
1783     n = MIN(n, s->sector_next_status - sector_num);
1784     if (s->status == BLK_DATA) {
1785         n = MIN(n, s->buf_sectors);
1786     }
1787 
1788     /* We need to write complete clusters for compressed images, so if an
1789      * unallocated area is shorter than that, we must consider the whole
1790      * cluster allocated. */
1791     if (s->compressed) {
1792         if (n < s->cluster_sectors) {
1793             n = MIN(s->cluster_sectors, s->total_sectors - sector_num);
1794             s->status = BLK_DATA;
1795         } else {
1796             n = QEMU_ALIGN_DOWN(n, s->cluster_sectors);
1797         }
1798     }
1799 
1800     return n;
1801 }
1802 
1803 static int coroutine_fn convert_co_read(ImgConvertState *s, int64_t sector_num,
1804                                         int nb_sectors, uint8_t *buf)
1805 {
1806     uint64_t single_read_until = 0;
1807     int n, ret;
1808 
1809     assert(nb_sectors <= s->buf_sectors);
1810     while (nb_sectors > 0) {
1811         BlockBackend *blk;
1812         int src_cur;
1813         int64_t bs_sectors, src_cur_offset;
1814         uint64_t offset;
1815 
1816         /* In the case of compression with multiple source files, we can get a
1817          * nb_sectors that spreads into the next part. So we must be able to
1818          * read across multiple BDSes for one convert_read() call. */
1819         convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1820         blk = s->src[src_cur];
1821         bs_sectors = s->src_sectors[src_cur];
1822 
1823         offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
1824 
1825         n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
1826         if (single_read_until > offset) {
1827             n = 1;
1828         }
1829 
1830         ret = blk_co_pread(blk, offset, n << BDRV_SECTOR_BITS, buf, 0);
1831         if (ret < 0) {
1832             if (s->salvage) {
1833                 if (n > 1) {
1834                     single_read_until = offset + (n << BDRV_SECTOR_BITS);
1835                     continue;
1836                 } else {
1837                     if (!s->quiet) {
1838                         warn_report("error while reading offset %" PRIu64
1839                                     ": %s", offset, strerror(-ret));
1840                     }
1841                     memset(buf, 0, BDRV_SECTOR_SIZE);
1842                 }
1843             } else {
1844                 return ret;
1845             }
1846         }
1847 
1848         sector_num += n;
1849         nb_sectors -= n;
1850         buf += n * BDRV_SECTOR_SIZE;
1851     }
1852 
1853     return 0;
1854 }
1855 
1856 
1857 static int coroutine_fn convert_co_write(ImgConvertState *s, int64_t sector_num,
1858                                          int nb_sectors, uint8_t *buf,
1859                                          enum ImgConvertBlockStatus status)
1860 {
1861     int ret;
1862 
1863     while (nb_sectors > 0) {
1864         int n = nb_sectors;
1865         BdrvRequestFlags flags = s->compressed ? BDRV_REQ_WRITE_COMPRESSED : 0;
1866 
1867         switch (status) {
1868         case BLK_BACKING_FILE:
1869             /* If we have a backing file, leave clusters unallocated that are
1870              * unallocated in the source image, so that the backing file is
1871              * visible at the respective offset. */
1872             assert(s->target_has_backing);
1873             break;
1874 
1875         case BLK_DATA:
1876             /* If we're told to keep the target fully allocated (-S 0) or there
1877              * is real non-zero data, we must write it. Otherwise we can treat
1878              * it as zero sectors.
1879              * Compressed clusters need to be written as a whole, so in that
1880              * case we can only save the write if the buffer is completely
1881              * zeroed. */
1882             if (!s->min_sparse ||
1883                 (!s->compressed &&
1884                  is_allocated_sectors_min(buf, n, &n, s->min_sparse,
1885                                           sector_num, s->alignment)) ||
1886                 (s->compressed &&
1887                  !buffer_is_zero(buf, n * BDRV_SECTOR_SIZE)))
1888             {
1889                 ret = blk_co_pwrite(s->target, sector_num << BDRV_SECTOR_BITS,
1890                                     n << BDRV_SECTOR_BITS, buf, flags);
1891                 if (ret < 0) {
1892                     return ret;
1893                 }
1894                 break;
1895             }
1896             /* fall-through */
1897 
1898         case BLK_ZERO:
1899             if (s->has_zero_init) {
1900                 assert(!s->target_has_backing);
1901                 break;
1902             }
1903             ret = blk_co_pwrite_zeroes(s->target,
1904                                        sector_num << BDRV_SECTOR_BITS,
1905                                        n << BDRV_SECTOR_BITS,
1906                                        BDRV_REQ_MAY_UNMAP);
1907             if (ret < 0) {
1908                 return ret;
1909             }
1910             break;
1911         }
1912 
1913         sector_num += n;
1914         nb_sectors -= n;
1915         buf += n * BDRV_SECTOR_SIZE;
1916     }
1917 
1918     return 0;
1919 }
1920 
1921 static int coroutine_fn convert_co_copy_range(ImgConvertState *s, int64_t sector_num,
1922                                               int nb_sectors)
1923 {
1924     int n, ret;
1925 
1926     while (nb_sectors > 0) {
1927         BlockBackend *blk;
1928         int src_cur;
1929         int64_t bs_sectors, src_cur_offset;
1930         int64_t offset;
1931 
1932         convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1933         offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
1934         blk = s->src[src_cur];
1935         bs_sectors = s->src_sectors[src_cur];
1936 
1937         n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
1938 
1939         ret = blk_co_copy_range(blk, offset, s->target,
1940                                 sector_num << BDRV_SECTOR_BITS,
1941                                 n << BDRV_SECTOR_BITS, 0, 0);
1942         if (ret < 0) {
1943             return ret;
1944         }
1945 
1946         sector_num += n;
1947         nb_sectors -= n;
1948     }
1949     return 0;
1950 }
1951 
1952 static void coroutine_fn convert_co_do_copy(void *opaque)
1953 {
1954     ImgConvertState *s = opaque;
1955     uint8_t *buf = NULL;
1956     int ret, i;
1957     int index = -1;
1958 
1959     for (i = 0; i < s->num_coroutines; i++) {
1960         if (s->co[i] == qemu_coroutine_self()) {
1961             index = i;
1962             break;
1963         }
1964     }
1965     assert(index >= 0);
1966 
1967     s->running_coroutines++;
1968     buf = blk_blockalign(s->target, s->buf_sectors * BDRV_SECTOR_SIZE);
1969 
1970     while (1) {
1971         int n;
1972         int64_t sector_num;
1973         enum ImgConvertBlockStatus status;
1974         bool copy_range;
1975 
1976         qemu_co_mutex_lock(&s->lock);
1977         if (s->ret != -EINPROGRESS || s->sector_num >= s->total_sectors) {
1978             qemu_co_mutex_unlock(&s->lock);
1979             break;
1980         }
1981         n = convert_iteration_sectors(s, s->sector_num);
1982         if (n < 0) {
1983             qemu_co_mutex_unlock(&s->lock);
1984             s->ret = n;
1985             break;
1986         }
1987         /* save current sector and allocation status to local variables */
1988         sector_num = s->sector_num;
1989         status = s->status;
1990         if (!s->min_sparse && s->status == BLK_ZERO) {
1991             n = MIN(n, s->buf_sectors);
1992         }
1993         /* increment global sector counter so that other coroutines can
1994          * already continue reading beyond this request */
1995         s->sector_num += n;
1996         qemu_co_mutex_unlock(&s->lock);
1997 
1998         if (status == BLK_DATA || (!s->min_sparse && status == BLK_ZERO)) {
1999             s->allocated_done += n;
2000             qemu_progress_print(100.0 * s->allocated_done /
2001                                         s->allocated_sectors, 0);
2002         }
2003 
2004 retry:
2005         copy_range = s->copy_range && s->status == BLK_DATA;
2006         if (status == BLK_DATA && !copy_range) {
2007             ret = convert_co_read(s, sector_num, n, buf);
2008             if (ret < 0) {
2009                 error_report("error while reading at byte %lld: %s",
2010                              sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
2011                 s->ret = ret;
2012             }
2013         } else if (!s->min_sparse && status == BLK_ZERO) {
2014             status = BLK_DATA;
2015             memset(buf, 0x00, n * BDRV_SECTOR_SIZE);
2016         }
2017 
2018         if (s->wr_in_order) {
2019             /* keep writes in order */
2020             while (s->wr_offs != sector_num && s->ret == -EINPROGRESS) {
2021                 s->wait_sector_num[index] = sector_num;
2022                 qemu_coroutine_yield();
2023             }
2024             s->wait_sector_num[index] = -1;
2025         }
2026 
2027         if (s->ret == -EINPROGRESS) {
2028             if (copy_range) {
2029                 ret = convert_co_copy_range(s, sector_num, n);
2030                 if (ret) {
2031                     s->copy_range = false;
2032                     goto retry;
2033                 }
2034             } else {
2035                 ret = convert_co_write(s, sector_num, n, buf, status);
2036             }
2037             if (ret < 0) {
2038                 error_report("error while writing at byte %lld: %s",
2039                              sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
2040                 s->ret = ret;
2041             }
2042         }
2043 
2044         if (s->wr_in_order) {
2045             /* reenter the coroutine that might have waited
2046              * for this write to complete */
2047             s->wr_offs = sector_num + n;
2048             for (i = 0; i < s->num_coroutines; i++) {
2049                 if (s->co[i] && s->wait_sector_num[i] == s->wr_offs) {
2050                     /*
2051                      * A -> B -> A cannot occur because A has
2052                      * s->wait_sector_num[i] == -1 during A -> B.  Therefore
2053                      * B will never enter A during this time window.
2054                      */
2055                     qemu_coroutine_enter(s->co[i]);
2056                     break;
2057                 }
2058             }
2059         }
2060     }
2061 
2062     qemu_vfree(buf);
2063     s->co[index] = NULL;
2064     s->running_coroutines--;
2065     if (!s->running_coroutines && s->ret == -EINPROGRESS) {
2066         /* the convert job finished successfully */
2067         s->ret = 0;
2068     }
2069 }
2070 
2071 static int convert_do_copy(ImgConvertState *s)
2072 {
2073     int ret, i, n;
2074     int64_t sector_num = 0;
2075 
2076     /* Check whether we have zero initialisation or can get it efficiently */
2077     if (!s->has_zero_init && s->target_is_new && s->min_sparse &&
2078         !s->target_has_backing) {
2079         s->has_zero_init = bdrv_has_zero_init(blk_bs(s->target));
2080     }
2081 
2082     /* Allocate buffer for copied data. For compressed images, only one cluster
2083      * can be copied at a time. */
2084     if (s->compressed) {
2085         if (s->cluster_sectors <= 0 || s->cluster_sectors > s->buf_sectors) {
2086             error_report("invalid cluster size");
2087             return -EINVAL;
2088         }
2089         s->buf_sectors = s->cluster_sectors;
2090     }
2091 
2092     while (sector_num < s->total_sectors) {
2093         n = convert_iteration_sectors(s, sector_num);
2094         if (n < 0) {
2095             return n;
2096         }
2097         if (s->status == BLK_DATA || (!s->min_sparse && s->status == BLK_ZERO))
2098         {
2099             s->allocated_sectors += n;
2100         }
2101         sector_num += n;
2102     }
2103 
2104     /* Do the copy */
2105     s->sector_next_status = 0;
2106     s->ret = -EINPROGRESS;
2107 
2108     qemu_co_mutex_init(&s->lock);
2109     for (i = 0; i < s->num_coroutines; i++) {
2110         s->co[i] = qemu_coroutine_create(convert_co_do_copy, s);
2111         s->wait_sector_num[i] = -1;
2112         qemu_coroutine_enter(s->co[i]);
2113     }
2114 
2115     while (s->running_coroutines) {
2116         main_loop_wait(false);
2117     }
2118 
2119     if (s->compressed && !s->ret) {
2120         /* signal EOF to align */
2121         ret = blk_pwrite_compressed(s->target, 0, 0, NULL);
2122         if (ret < 0) {
2123             return ret;
2124         }
2125     }
2126 
2127     return s->ret;
2128 }
2129 
2130 /* Check that bitmaps can be copied, or output an error */
2131 static int convert_check_bitmaps(BlockDriverState *src, bool skip_broken)
2132 {
2133     BdrvDirtyBitmap *bm;
2134 
2135     if (!bdrv_supports_persistent_dirty_bitmap(src)) {
2136         error_report("Source lacks bitmap support");
2137         return -1;
2138     }
2139     FOR_EACH_DIRTY_BITMAP(src, bm) {
2140         if (!bdrv_dirty_bitmap_get_persistence(bm)) {
2141             continue;
2142         }
2143         if (!skip_broken && bdrv_dirty_bitmap_inconsistent(bm)) {
2144             error_report("Cannot copy inconsistent bitmap '%s'",
2145                          bdrv_dirty_bitmap_name(bm));
2146             error_printf("Try --skip-broken-bitmaps, or "
2147                          "use 'qemu-img bitmap --remove' to delete it\n");
2148             return -1;
2149         }
2150     }
2151     return 0;
2152 }
2153 
2154 static int convert_copy_bitmaps(BlockDriverState *src, BlockDriverState *dst,
2155                                 bool skip_broken)
2156 {
2157     BdrvDirtyBitmap *bm;
2158     Error *err = NULL;
2159 
2160     FOR_EACH_DIRTY_BITMAP(src, bm) {
2161         const char *name;
2162 
2163         if (!bdrv_dirty_bitmap_get_persistence(bm)) {
2164             continue;
2165         }
2166         name = bdrv_dirty_bitmap_name(bm);
2167         if (skip_broken && bdrv_dirty_bitmap_inconsistent(bm)) {
2168             warn_report("Skipping inconsistent bitmap '%s'", name);
2169             continue;
2170         }
2171         qmp_block_dirty_bitmap_add(dst->node_name, name,
2172                                    true, bdrv_dirty_bitmap_granularity(bm),
2173                                    true, true,
2174                                    true, !bdrv_dirty_bitmap_enabled(bm),
2175                                    &err);
2176         if (err) {
2177             error_reportf_err(err, "Failed to create bitmap %s: ", name);
2178             return -1;
2179         }
2180 
2181         do_dirty_bitmap_merge(dst->node_name, name, src->node_name, name,
2182                               &err);
2183         if (err) {
2184             error_reportf_err(err, "Failed to populate bitmap %s: ", name);
2185             qmp_block_dirty_bitmap_remove(dst->node_name, name, NULL);
2186             return -1;
2187         }
2188     }
2189 
2190     return 0;
2191 }
2192 
2193 #define MAX_BUF_SECTORS 32768
2194 
2195 static void set_rate_limit(BlockBackend *blk, int64_t rate_limit)
2196 {
2197     ThrottleConfig cfg;
2198 
2199     throttle_config_init(&cfg);
2200     cfg.buckets[THROTTLE_BPS_WRITE].avg = rate_limit;
2201 
2202     blk_io_limits_enable(blk, CONVERT_THROTTLE_GROUP);
2203     blk_set_io_limits(blk, &cfg);
2204 }
2205 
2206 static int img_convert(int argc, char **argv)
2207 {
2208     int c, bs_i, flags, src_flags = BDRV_O_NO_SHARE;
2209     const char *fmt = NULL, *out_fmt = NULL, *cache = "unsafe",
2210                *src_cache = BDRV_DEFAULT_CACHE, *out_baseimg = NULL,
2211                *out_filename, *out_baseimg_param, *snapshot_name = NULL,
2212                *backing_fmt = NULL;
2213     BlockDriver *drv = NULL, *proto_drv = NULL;
2214     BlockDriverInfo bdi;
2215     BlockDriverState *out_bs;
2216     QemuOpts *opts = NULL, *sn_opts = NULL;
2217     QemuOptsList *create_opts = NULL;
2218     QDict *open_opts = NULL;
2219     char *options = NULL;
2220     Error *local_err = NULL;
2221     bool writethrough, src_writethrough, image_opts = false,
2222          skip_create = false, progress = false, tgt_image_opts = false;
2223     int64_t ret = -EINVAL;
2224     bool force_share = false;
2225     bool explict_min_sparse = false;
2226     bool bitmaps = false;
2227     bool skip_broken = false;
2228     int64_t rate_limit = 0;
2229 
2230     ImgConvertState s = (ImgConvertState) {
2231         /* Need at least 4k of zeros for sparse detection */
2232         .min_sparse         = 8,
2233         .copy_range         = false,
2234         .buf_sectors        = IO_BUF_SIZE / BDRV_SECTOR_SIZE,
2235         .wr_in_order        = true,
2236         .num_coroutines     = 8,
2237     };
2238 
2239     for(;;) {
2240         static const struct option long_options[] = {
2241             {"help", no_argument, 0, 'h'},
2242             {"object", required_argument, 0, OPTION_OBJECT},
2243             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2244             {"force-share", no_argument, 0, 'U'},
2245             {"target-image-opts", no_argument, 0, OPTION_TARGET_IMAGE_OPTS},
2246             {"salvage", no_argument, 0, OPTION_SALVAGE},
2247             {"target-is-zero", no_argument, 0, OPTION_TARGET_IS_ZERO},
2248             {"bitmaps", no_argument, 0, OPTION_BITMAPS},
2249             {"skip-broken-bitmaps", no_argument, 0, OPTION_SKIP_BROKEN},
2250             {0, 0, 0, 0}
2251         };
2252         c = getopt_long(argc, argv, ":hf:O:B:CcF:o:l:S:pt:T:qnm:WUr:",
2253                         long_options, NULL);
2254         if (c == -1) {
2255             break;
2256         }
2257         switch(c) {
2258         case ':':
2259             missing_argument(argv[optind - 1]);
2260             break;
2261         case '?':
2262             unrecognized_option(argv[optind - 1]);
2263             break;
2264         case 'h':
2265             help();
2266             break;
2267         case 'f':
2268             fmt = optarg;
2269             break;
2270         case 'O':
2271             out_fmt = optarg;
2272             break;
2273         case 'B':
2274             out_baseimg = optarg;
2275             break;
2276         case 'C':
2277             s.copy_range = true;
2278             break;
2279         case 'c':
2280             s.compressed = true;
2281             break;
2282         case 'F':
2283             backing_fmt = optarg;
2284             break;
2285         case 'o':
2286             if (accumulate_options(&options, optarg) < 0) {
2287                 goto fail_getopt;
2288             }
2289             break;
2290         case 'l':
2291             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
2292                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
2293                                                   optarg, false);
2294                 if (!sn_opts) {
2295                     error_report("Failed in parsing snapshot param '%s'",
2296                                  optarg);
2297                     goto fail_getopt;
2298                 }
2299             } else {
2300                 snapshot_name = optarg;
2301             }
2302             break;
2303         case 'S':
2304         {
2305             int64_t sval;
2306 
2307             sval = cvtnum("buffer size for sparse output", optarg);
2308             if (sval < 0) {
2309                 goto fail_getopt;
2310             } else if (!QEMU_IS_ALIGNED(sval, BDRV_SECTOR_SIZE) ||
2311                 sval / BDRV_SECTOR_SIZE > MAX_BUF_SECTORS) {
2312                 error_report("Invalid buffer size for sparse output specified. "
2313                     "Valid sizes are multiples of %llu up to %llu. Select "
2314                     "0 to disable sparse detection (fully allocates output).",
2315                     BDRV_SECTOR_SIZE, MAX_BUF_SECTORS * BDRV_SECTOR_SIZE);
2316                 goto fail_getopt;
2317             }
2318 
2319             s.min_sparse = sval / BDRV_SECTOR_SIZE;
2320             explict_min_sparse = true;
2321             break;
2322         }
2323         case 'p':
2324             progress = true;
2325             break;
2326         case 't':
2327             cache = optarg;
2328             break;
2329         case 'T':
2330             src_cache = optarg;
2331             break;
2332         case 'q':
2333             s.quiet = true;
2334             break;
2335         case 'n':
2336             skip_create = true;
2337             break;
2338         case 'm':
2339             if (qemu_strtol(optarg, NULL, 0, &s.num_coroutines) ||
2340                 s.num_coroutines < 1 || s.num_coroutines > MAX_COROUTINES) {
2341                 error_report("Invalid number of coroutines. Allowed number of"
2342                              " coroutines is between 1 and %d", MAX_COROUTINES);
2343                 goto fail_getopt;
2344             }
2345             break;
2346         case 'W':
2347             s.wr_in_order = false;
2348             break;
2349         case 'U':
2350             force_share = true;
2351             break;
2352         case 'r':
2353             rate_limit = cvtnum("rate limit", optarg);
2354             if (rate_limit < 0) {
2355                 goto fail_getopt;
2356             }
2357             break;
2358         case OPTION_OBJECT:
2359             user_creatable_process_cmdline(optarg);
2360             break;
2361         case OPTION_IMAGE_OPTS:
2362             image_opts = true;
2363             break;
2364         case OPTION_SALVAGE:
2365             s.salvage = true;
2366             break;
2367         case OPTION_TARGET_IMAGE_OPTS:
2368             tgt_image_opts = true;
2369             break;
2370         case OPTION_TARGET_IS_ZERO:
2371             /*
2372              * The user asserting that the target is blank has the
2373              * same effect as the target driver supporting zero
2374              * initialisation.
2375              */
2376             s.has_zero_init = true;
2377             break;
2378         case OPTION_BITMAPS:
2379             bitmaps = true;
2380             break;
2381         case OPTION_SKIP_BROKEN:
2382             skip_broken = true;
2383             break;
2384         }
2385     }
2386 
2387     if (!out_fmt && !tgt_image_opts) {
2388         out_fmt = "raw";
2389     }
2390 
2391     if (skip_broken && !bitmaps) {
2392         error_report("Use of --skip-broken-bitmaps requires --bitmaps");
2393         goto fail_getopt;
2394     }
2395 
2396     if (s.compressed && s.copy_range) {
2397         error_report("Cannot enable copy offloading when -c is used");
2398         goto fail_getopt;
2399     }
2400 
2401     if (explict_min_sparse && s.copy_range) {
2402         error_report("Cannot enable copy offloading when -S is used");
2403         goto fail_getopt;
2404     }
2405 
2406     if (s.copy_range && s.salvage) {
2407         error_report("Cannot use copy offloading in salvaging mode");
2408         goto fail_getopt;
2409     }
2410 
2411     if (tgt_image_opts && !skip_create) {
2412         error_report("--target-image-opts requires use of -n flag");
2413         goto fail_getopt;
2414     }
2415 
2416     if (skip_create && options) {
2417         error_report("-o has no effect when skipping image creation");
2418         goto fail_getopt;
2419     }
2420 
2421     if (s.has_zero_init && !skip_create) {
2422         error_report("--target-is-zero requires use of -n flag");
2423         goto fail_getopt;
2424     }
2425 
2426     s.src_num = argc - optind - 1;
2427     out_filename = s.src_num >= 1 ? argv[argc - 1] : NULL;
2428 
2429     if (options && has_help_option(options)) {
2430         if (out_fmt) {
2431             ret = print_block_option_help(out_filename, out_fmt);
2432             goto fail_getopt;
2433         } else {
2434             error_report("Option help requires a format be specified");
2435             goto fail_getopt;
2436         }
2437     }
2438 
2439     if (s.src_num < 1) {
2440         error_report("Must specify image file name");
2441         goto fail_getopt;
2442     }
2443 
2444     /* ret is still -EINVAL until here */
2445     ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
2446     if (ret < 0) {
2447         error_report("Invalid source cache option: %s", src_cache);
2448         goto fail_getopt;
2449     }
2450 
2451     /* Initialize before goto out */
2452     if (s.quiet) {
2453         progress = false;
2454     }
2455     qemu_progress_init(progress, 1.0);
2456     qemu_progress_print(0, 100);
2457 
2458     s.src = g_new0(BlockBackend *, s.src_num);
2459     s.src_sectors = g_new(int64_t, s.src_num);
2460     s.src_alignment = g_new(int, s.src_num);
2461 
2462     for (bs_i = 0; bs_i < s.src_num; bs_i++) {
2463         BlockDriverState *src_bs;
2464         s.src[bs_i] = img_open(image_opts, argv[optind + bs_i],
2465                                fmt, src_flags, src_writethrough, s.quiet,
2466                                force_share);
2467         if (!s.src[bs_i]) {
2468             ret = -1;
2469             goto out;
2470         }
2471         s.src_sectors[bs_i] = blk_nb_sectors(s.src[bs_i]);
2472         if (s.src_sectors[bs_i] < 0) {
2473             error_report("Could not get size of %s: %s",
2474                          argv[optind + bs_i], strerror(-s.src_sectors[bs_i]));
2475             ret = -1;
2476             goto out;
2477         }
2478         src_bs = blk_bs(s.src[bs_i]);
2479         s.src_alignment[bs_i] = DIV_ROUND_UP(src_bs->bl.request_alignment,
2480                                              BDRV_SECTOR_SIZE);
2481         if (!bdrv_get_info(src_bs, &bdi)) {
2482             s.src_alignment[bs_i] = MAX(s.src_alignment[bs_i],
2483                                         bdi.cluster_size / BDRV_SECTOR_SIZE);
2484         }
2485         s.total_sectors += s.src_sectors[bs_i];
2486     }
2487 
2488     if (sn_opts) {
2489         bdrv_snapshot_load_tmp(blk_bs(s.src[0]),
2490                                qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
2491                                qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
2492                                &local_err);
2493     } else if (snapshot_name != NULL) {
2494         if (s.src_num > 1) {
2495             error_report("No support for concatenating multiple snapshot");
2496             ret = -1;
2497             goto out;
2498         }
2499 
2500         bdrv_snapshot_load_tmp_by_id_or_name(blk_bs(s.src[0]), snapshot_name,
2501                                              &local_err);
2502     }
2503     if (local_err) {
2504         error_reportf_err(local_err, "Failed to load snapshot: ");
2505         ret = -1;
2506         goto out;
2507     }
2508 
2509     if (!skip_create) {
2510         /* Find driver and parse its options */
2511         drv = bdrv_find_format(out_fmt);
2512         if (!drv) {
2513             error_report("Unknown file format '%s'", out_fmt);
2514             ret = -1;
2515             goto out;
2516         }
2517 
2518         proto_drv = bdrv_find_protocol(out_filename, true, &local_err);
2519         if (!proto_drv) {
2520             error_report_err(local_err);
2521             ret = -1;
2522             goto out;
2523         }
2524 
2525         if (!drv->create_opts) {
2526             error_report("Format driver '%s' does not support image creation",
2527                          drv->format_name);
2528             ret = -1;
2529             goto out;
2530         }
2531 
2532         if (!proto_drv->create_opts) {
2533             error_report("Protocol driver '%s' does not support image creation",
2534                          proto_drv->format_name);
2535             ret = -1;
2536             goto out;
2537         }
2538 
2539         create_opts = qemu_opts_append(create_opts, drv->create_opts);
2540         create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
2541 
2542         opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
2543         if (options) {
2544             if (!qemu_opts_do_parse(opts, options, NULL, &local_err)) {
2545                 error_report_err(local_err);
2546                 ret = -1;
2547                 goto out;
2548             }
2549         }
2550 
2551         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2552                             s.total_sectors * BDRV_SECTOR_SIZE, &error_abort);
2553         ret = add_old_style_options(out_fmt, opts, out_baseimg, backing_fmt);
2554         if (ret < 0) {
2555             goto out;
2556         }
2557     }
2558 
2559     /* Get backing file name if -o backing_file was used */
2560     out_baseimg_param = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
2561     if (out_baseimg_param) {
2562         out_baseimg = out_baseimg_param;
2563     }
2564     s.target_has_backing = (bool) out_baseimg;
2565 
2566     if (s.has_zero_init && s.target_has_backing) {
2567         error_report("Cannot use --target-is-zero when the destination "
2568                      "image has a backing file");
2569         goto out;
2570     }
2571 
2572     if (s.src_num > 1 && out_baseimg) {
2573         error_report("Having a backing file for the target makes no sense when "
2574                      "concatenating multiple input images");
2575         ret = -1;
2576         goto out;
2577     }
2578 
2579     if (out_baseimg_param) {
2580         if (!qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT)) {
2581             error_report("Use of backing file requires explicit "
2582                          "backing format");
2583             ret = -1;
2584             goto out;
2585         }
2586     }
2587 
2588     /* Check if compression is supported */
2589     if (s.compressed) {
2590         bool encryption =
2591             qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, false);
2592         const char *encryptfmt =
2593             qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT);
2594         const char *preallocation =
2595             qemu_opt_get(opts, BLOCK_OPT_PREALLOC);
2596 
2597         if (drv && !block_driver_can_compress(drv)) {
2598             error_report("Compression not supported for this file format");
2599             ret = -1;
2600             goto out;
2601         }
2602 
2603         if (encryption || encryptfmt) {
2604             error_report("Compression and encryption not supported at "
2605                          "the same time");
2606             ret = -1;
2607             goto out;
2608         }
2609 
2610         if (preallocation
2611             && strcmp(preallocation, "off"))
2612         {
2613             error_report("Compression and preallocation not supported at "
2614                          "the same time");
2615             ret = -1;
2616             goto out;
2617         }
2618     }
2619 
2620     /* Determine if bitmaps need copying */
2621     if (bitmaps) {
2622         if (s.src_num > 1) {
2623             error_report("Copying bitmaps only possible with single source");
2624             ret = -1;
2625             goto out;
2626         }
2627         ret = convert_check_bitmaps(blk_bs(s.src[0]), skip_broken);
2628         if (ret < 0) {
2629             goto out;
2630         }
2631     }
2632 
2633     /*
2634      * The later open call will need any decryption secrets, and
2635      * bdrv_create() will purge "opts", so extract them now before
2636      * they are lost.
2637      */
2638     if (!skip_create) {
2639         open_opts = qdict_new();
2640         qemu_opt_foreach(opts, img_add_key_secrets, open_opts, &error_abort);
2641 
2642         /* Create the new image */
2643         ret = bdrv_create(drv, out_filename, opts, &local_err);
2644         if (ret < 0) {
2645             error_reportf_err(local_err, "%s: error while converting %s: ",
2646                               out_filename, out_fmt);
2647             goto out;
2648         }
2649     }
2650 
2651     s.target_is_new = !skip_create;
2652 
2653     flags = s.min_sparse ? (BDRV_O_RDWR | BDRV_O_UNMAP) : BDRV_O_RDWR;
2654     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
2655     if (ret < 0) {
2656         error_report("Invalid cache option: %s", cache);
2657         goto out;
2658     }
2659 
2660     if (flags & BDRV_O_NOCACHE) {
2661         /*
2662          * If we open the target with O_DIRECT, it may be necessary to
2663          * extend its size to align to the physical sector size.
2664          */
2665         flags |= BDRV_O_RESIZE;
2666     }
2667 
2668     if (skip_create) {
2669         s.target = img_open(tgt_image_opts, out_filename, out_fmt,
2670                             flags, writethrough, s.quiet, false);
2671     } else {
2672         /* TODO ultimately we should allow --target-image-opts
2673          * to be used even when -n is not given.
2674          * That has to wait for bdrv_create to be improved
2675          * to allow filenames in option syntax
2676          */
2677         s.target = img_open_file(out_filename, open_opts, out_fmt,
2678                                  flags, writethrough, s.quiet, false);
2679         open_opts = NULL; /* blk_new_open will have freed it */
2680     }
2681     if (!s.target) {
2682         ret = -1;
2683         goto out;
2684     }
2685     out_bs = blk_bs(s.target);
2686 
2687     if (bitmaps && !bdrv_supports_persistent_dirty_bitmap(out_bs)) {
2688         error_report("Format driver '%s' does not support bitmaps",
2689                      out_bs->drv->format_name);
2690         ret = -1;
2691         goto out;
2692     }
2693 
2694     if (s.compressed && !block_driver_can_compress(out_bs->drv)) {
2695         error_report("Compression not supported for this file format");
2696         ret = -1;
2697         goto out;
2698     }
2699 
2700     /* increase bufsectors from the default 4096 (2M) if opt_transfer
2701      * or discard_alignment of the out_bs is greater. Limit to
2702      * MAX_BUF_SECTORS as maximum which is currently 32768 (16MB). */
2703     s.buf_sectors = MIN(MAX_BUF_SECTORS,
2704                         MAX(s.buf_sectors,
2705                             MAX(out_bs->bl.opt_transfer >> BDRV_SECTOR_BITS,
2706                                 out_bs->bl.pdiscard_alignment >>
2707                                 BDRV_SECTOR_BITS)));
2708 
2709     /* try to align the write requests to the destination to avoid unnecessary
2710      * RMW cycles. */
2711     s.alignment = MAX(pow2floor(s.min_sparse),
2712                       DIV_ROUND_UP(out_bs->bl.request_alignment,
2713                                    BDRV_SECTOR_SIZE));
2714     assert(is_power_of_2(s.alignment));
2715 
2716     if (skip_create) {
2717         int64_t output_sectors = blk_nb_sectors(s.target);
2718         if (output_sectors < 0) {
2719             error_report("unable to get output image length: %s",
2720                          strerror(-output_sectors));
2721             ret = -1;
2722             goto out;
2723         } else if (output_sectors < s.total_sectors) {
2724             error_report("output file is smaller than input file");
2725             ret = -1;
2726             goto out;
2727         }
2728     }
2729 
2730     if (s.target_has_backing && s.target_is_new) {
2731         /* Errors are treated as "backing length unknown" (which means
2732          * s.target_backing_sectors has to be negative, which it will
2733          * be automatically).  The backing file length is used only
2734          * for optimizations, so such a case is not fatal. */
2735         s.target_backing_sectors =
2736             bdrv_nb_sectors(bdrv_backing_chain_next(out_bs));
2737     } else {
2738         s.target_backing_sectors = -1;
2739     }
2740 
2741     ret = bdrv_get_info(out_bs, &bdi);
2742     if (ret < 0) {
2743         if (s.compressed) {
2744             error_report("could not get block driver info");
2745             goto out;
2746         }
2747     } else {
2748         s.compressed = s.compressed || bdi.needs_compressed_writes;
2749         s.cluster_sectors = bdi.cluster_size / BDRV_SECTOR_SIZE;
2750     }
2751 
2752     if (rate_limit) {
2753         set_rate_limit(s.target, rate_limit);
2754     }
2755 
2756     ret = convert_do_copy(&s);
2757 
2758     /* Now copy the bitmaps */
2759     if (bitmaps && ret == 0) {
2760         ret = convert_copy_bitmaps(blk_bs(s.src[0]), out_bs, skip_broken);
2761     }
2762 
2763 out:
2764     if (!ret) {
2765         qemu_progress_print(100, 0);
2766     }
2767     qemu_progress_end();
2768     qemu_opts_del(opts);
2769     qemu_opts_free(create_opts);
2770     qobject_unref(open_opts);
2771     blk_unref(s.target);
2772     if (s.src) {
2773         for (bs_i = 0; bs_i < s.src_num; bs_i++) {
2774             blk_unref(s.src[bs_i]);
2775         }
2776         g_free(s.src);
2777     }
2778     g_free(s.src_sectors);
2779     g_free(s.src_alignment);
2780 fail_getopt:
2781     qemu_opts_del(sn_opts);
2782     g_free(options);
2783 
2784     return !!ret;
2785 }
2786 
2787 
2788 static void dump_snapshots(BlockDriverState *bs)
2789 {
2790     QEMUSnapshotInfo *sn_tab, *sn;
2791     int nb_sns, i;
2792 
2793     nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2794     if (nb_sns <= 0)
2795         return;
2796     printf("Snapshot list:\n");
2797     bdrv_snapshot_dump(NULL);
2798     printf("\n");
2799     for(i = 0; i < nb_sns; i++) {
2800         sn = &sn_tab[i];
2801         bdrv_snapshot_dump(sn);
2802         printf("\n");
2803     }
2804     g_free(sn_tab);
2805 }
2806 
2807 static void dump_json_image_info_list(ImageInfoList *list)
2808 {
2809     GString *str;
2810     QObject *obj;
2811     Visitor *v = qobject_output_visitor_new(&obj);
2812 
2813     visit_type_ImageInfoList(v, NULL, &list, &error_abort);
2814     visit_complete(v, &obj);
2815     str = qobject_to_json_pretty(obj, true);
2816     assert(str != NULL);
2817     printf("%s\n", str->str);
2818     qobject_unref(obj);
2819     visit_free(v);
2820     g_string_free(str, true);
2821 }
2822 
2823 static void dump_json_image_info(ImageInfo *info)
2824 {
2825     GString *str;
2826     QObject *obj;
2827     Visitor *v = qobject_output_visitor_new(&obj);
2828 
2829     visit_type_ImageInfo(v, NULL, &info, &error_abort);
2830     visit_complete(v, &obj);
2831     str = qobject_to_json_pretty(obj, true);
2832     assert(str != NULL);
2833     printf("%s\n", str->str);
2834     qobject_unref(obj);
2835     visit_free(v);
2836     g_string_free(str, true);
2837 }
2838 
2839 static void dump_human_image_info_list(ImageInfoList *list)
2840 {
2841     ImageInfoList *elem;
2842     bool delim = false;
2843 
2844     for (elem = list; elem; elem = elem->next) {
2845         if (delim) {
2846             printf("\n");
2847         }
2848         delim = true;
2849 
2850         bdrv_image_info_dump(elem->value);
2851     }
2852 }
2853 
2854 static gboolean str_equal_func(gconstpointer a, gconstpointer b)
2855 {
2856     return strcmp(a, b) == 0;
2857 }
2858 
2859 /**
2860  * Open an image file chain and return an ImageInfoList
2861  *
2862  * @filename: topmost image filename
2863  * @fmt: topmost image format (may be NULL to autodetect)
2864  * @chain: true  - enumerate entire backing file chain
2865  *         false - only topmost image file
2866  *
2867  * Returns a list of ImageInfo objects or NULL if there was an error opening an
2868  * image file.  If there was an error a message will have been printed to
2869  * stderr.
2870  */
2871 static ImageInfoList *collect_image_info_list(bool image_opts,
2872                                               const char *filename,
2873                                               const char *fmt,
2874                                               bool chain, bool force_share)
2875 {
2876     ImageInfoList *head = NULL;
2877     ImageInfoList **tail = &head;
2878     GHashTable *filenames;
2879     Error *err = NULL;
2880 
2881     filenames = g_hash_table_new_full(g_str_hash, str_equal_func, NULL, NULL);
2882 
2883     while (filename) {
2884         BlockBackend *blk;
2885         BlockDriverState *bs;
2886         ImageInfo *info;
2887 
2888         if (g_hash_table_lookup_extended(filenames, filename, NULL, NULL)) {
2889             error_report("Backing file '%s' creates an infinite loop.",
2890                          filename);
2891             goto err;
2892         }
2893         g_hash_table_insert(filenames, (gpointer)filename, NULL);
2894 
2895         blk = img_open(image_opts, filename, fmt,
2896                        BDRV_O_NO_BACKING | BDRV_O_NO_IO, false, false,
2897                        force_share);
2898         if (!blk) {
2899             goto err;
2900         }
2901         bs = blk_bs(blk);
2902 
2903         bdrv_query_image_info(bs, &info, &err);
2904         if (err) {
2905             error_report_err(err);
2906             blk_unref(blk);
2907             goto err;
2908         }
2909 
2910         QAPI_LIST_APPEND(tail, info);
2911 
2912         blk_unref(blk);
2913 
2914         /* Clear parameters that only apply to the topmost image */
2915         filename = fmt = NULL;
2916         image_opts = false;
2917 
2918         if (chain) {
2919             if (info->full_backing_filename) {
2920                 filename = info->full_backing_filename;
2921             } else if (info->backing_filename) {
2922                 error_report("Could not determine absolute backing filename,"
2923                              " but backing filename '%s' present",
2924                              info->backing_filename);
2925                 goto err;
2926             }
2927             if (info->backing_filename_format) {
2928                 fmt = info->backing_filename_format;
2929             }
2930         }
2931     }
2932     g_hash_table_destroy(filenames);
2933     return head;
2934 
2935 err:
2936     qapi_free_ImageInfoList(head);
2937     g_hash_table_destroy(filenames);
2938     return NULL;
2939 }
2940 
2941 static int img_info(int argc, char **argv)
2942 {
2943     int c;
2944     OutputFormat output_format = OFORMAT_HUMAN;
2945     bool chain = false;
2946     const char *filename, *fmt, *output;
2947     ImageInfoList *list;
2948     bool image_opts = false;
2949     bool force_share = false;
2950 
2951     fmt = NULL;
2952     output = NULL;
2953     for(;;) {
2954         int option_index = 0;
2955         static const struct option long_options[] = {
2956             {"help", no_argument, 0, 'h'},
2957             {"format", required_argument, 0, 'f'},
2958             {"output", required_argument, 0, OPTION_OUTPUT},
2959             {"backing-chain", no_argument, 0, OPTION_BACKING_CHAIN},
2960             {"object", required_argument, 0, OPTION_OBJECT},
2961             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2962             {"force-share", no_argument, 0, 'U'},
2963             {0, 0, 0, 0}
2964         };
2965         c = getopt_long(argc, argv, ":f:hU",
2966                         long_options, &option_index);
2967         if (c == -1) {
2968             break;
2969         }
2970         switch(c) {
2971         case ':':
2972             missing_argument(argv[optind - 1]);
2973             break;
2974         case '?':
2975             unrecognized_option(argv[optind - 1]);
2976             break;
2977         case 'h':
2978             help();
2979             break;
2980         case 'f':
2981             fmt = optarg;
2982             break;
2983         case 'U':
2984             force_share = true;
2985             break;
2986         case OPTION_OUTPUT:
2987             output = optarg;
2988             break;
2989         case OPTION_BACKING_CHAIN:
2990             chain = true;
2991             break;
2992         case OPTION_OBJECT:
2993             user_creatable_process_cmdline(optarg);
2994             break;
2995         case OPTION_IMAGE_OPTS:
2996             image_opts = true;
2997             break;
2998         }
2999     }
3000     if (optind != argc - 1) {
3001         error_exit("Expecting one image file name");
3002     }
3003     filename = argv[optind++];
3004 
3005     if (output && !strcmp(output, "json")) {
3006         output_format = OFORMAT_JSON;
3007     } else if (output && !strcmp(output, "human")) {
3008         output_format = OFORMAT_HUMAN;
3009     } else if (output) {
3010         error_report("--output must be used with human or json as argument.");
3011         return 1;
3012     }
3013 
3014     list = collect_image_info_list(image_opts, filename, fmt, chain,
3015                                    force_share);
3016     if (!list) {
3017         return 1;
3018     }
3019 
3020     switch (output_format) {
3021     case OFORMAT_HUMAN:
3022         dump_human_image_info_list(list);
3023         break;
3024     case OFORMAT_JSON:
3025         if (chain) {
3026             dump_json_image_info_list(list);
3027         } else {
3028             dump_json_image_info(list->value);
3029         }
3030         break;
3031     }
3032 
3033     qapi_free_ImageInfoList(list);
3034     return 0;
3035 }
3036 
3037 static int dump_map_entry(OutputFormat output_format, MapEntry *e,
3038                           MapEntry *next)
3039 {
3040     switch (output_format) {
3041     case OFORMAT_HUMAN:
3042         if (e->data && !e->has_offset) {
3043             error_report("File contains external, encrypted or compressed clusters.");
3044             return -1;
3045         }
3046         if (e->data && !e->zero) {
3047             printf("%#-16"PRIx64"%#-16"PRIx64"%#-16"PRIx64"%s\n",
3048                    e->start, e->length,
3049                    e->has_offset ? e->offset : 0,
3050                    e->filename ?: "");
3051         }
3052         /* This format ignores the distinction between 0, ZERO and ZERO|DATA.
3053          * Modify the flags here to allow more coalescing.
3054          */
3055         if (next && (!next->data || next->zero)) {
3056             next->data = false;
3057             next->zero = true;
3058         }
3059         break;
3060     case OFORMAT_JSON:
3061         printf("{ \"start\": %"PRId64", \"length\": %"PRId64","
3062                " \"depth\": %"PRId64", \"present\": %s, \"zero\": %s,"
3063                " \"data\": %s", e->start, e->length, e->depth,
3064                e->present ? "true" : "false",
3065                e->zero ? "true" : "false",
3066                e->data ? "true" : "false");
3067         if (e->has_offset) {
3068             printf(", \"offset\": %"PRId64"", e->offset);
3069         }
3070         putchar('}');
3071 
3072         if (next) {
3073             puts(",");
3074         }
3075         break;
3076     }
3077     return 0;
3078 }
3079 
3080 static int get_block_status(BlockDriverState *bs, int64_t offset,
3081                             int64_t bytes, MapEntry *e)
3082 {
3083     int ret;
3084     int depth;
3085     BlockDriverState *file;
3086     bool has_offset;
3087     int64_t map;
3088     char *filename = NULL;
3089 
3090     /* As an optimization, we could cache the current range of unallocated
3091      * clusters in each file of the chain, and avoid querying the same
3092      * range repeatedly.
3093      */
3094 
3095     depth = 0;
3096     for (;;) {
3097         bs = bdrv_skip_filters(bs);
3098         ret = bdrv_block_status(bs, offset, bytes, &bytes, &map, &file);
3099         if (ret < 0) {
3100             return ret;
3101         }
3102         assert(bytes);
3103         if (ret & (BDRV_BLOCK_ZERO|BDRV_BLOCK_DATA)) {
3104             break;
3105         }
3106         bs = bdrv_cow_bs(bs);
3107         if (bs == NULL) {
3108             ret = 0;
3109             break;
3110         }
3111 
3112         depth++;
3113     }
3114 
3115     has_offset = !!(ret & BDRV_BLOCK_OFFSET_VALID);
3116 
3117     if (file && has_offset) {
3118         bdrv_refresh_filename(file);
3119         filename = file->filename;
3120     }
3121 
3122     *e = (MapEntry) {
3123         .start = offset,
3124         .length = bytes,
3125         .data = !!(ret & BDRV_BLOCK_DATA),
3126         .zero = !!(ret & BDRV_BLOCK_ZERO),
3127         .offset = map,
3128         .has_offset = has_offset,
3129         .depth = depth,
3130         .present = !!(ret & BDRV_BLOCK_ALLOCATED),
3131         .filename = filename,
3132     };
3133 
3134     return 0;
3135 }
3136 
3137 static inline bool entry_mergeable(const MapEntry *curr, const MapEntry *next)
3138 {
3139     if (curr->length == 0) {
3140         return false;
3141     }
3142     if (curr->zero != next->zero ||
3143         curr->data != next->data ||
3144         curr->depth != next->depth ||
3145         curr->present != next->present ||
3146         !curr->filename != !next->filename ||
3147         curr->has_offset != next->has_offset) {
3148         return false;
3149     }
3150     if (curr->filename && strcmp(curr->filename, next->filename)) {
3151         return false;
3152     }
3153     if (curr->has_offset && curr->offset + curr->length != next->offset) {
3154         return false;
3155     }
3156     return true;
3157 }
3158 
3159 static int img_map(int argc, char **argv)
3160 {
3161     int c;
3162     OutputFormat output_format = OFORMAT_HUMAN;
3163     BlockBackend *blk;
3164     BlockDriverState *bs;
3165     const char *filename, *fmt, *output;
3166     int64_t length;
3167     MapEntry curr = { .length = 0 }, next;
3168     int ret = 0;
3169     bool image_opts = false;
3170     bool force_share = false;
3171     int64_t start_offset = 0;
3172     int64_t max_length = -1;
3173 
3174     fmt = NULL;
3175     output = NULL;
3176     for (;;) {
3177         int option_index = 0;
3178         static const struct option long_options[] = {
3179             {"help", no_argument, 0, 'h'},
3180             {"format", required_argument, 0, 'f'},
3181             {"output", required_argument, 0, OPTION_OUTPUT},
3182             {"object", required_argument, 0, OPTION_OBJECT},
3183             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3184             {"force-share", no_argument, 0, 'U'},
3185             {"start-offset", required_argument, 0, 's'},
3186             {"max-length", required_argument, 0, 'l'},
3187             {0, 0, 0, 0}
3188         };
3189         c = getopt_long(argc, argv, ":f:s:l:hU",
3190                         long_options, &option_index);
3191         if (c == -1) {
3192             break;
3193         }
3194         switch (c) {
3195         case ':':
3196             missing_argument(argv[optind - 1]);
3197             break;
3198         case '?':
3199             unrecognized_option(argv[optind - 1]);
3200             break;
3201         case 'h':
3202             help();
3203             break;
3204         case 'f':
3205             fmt = optarg;
3206             break;
3207         case 'U':
3208             force_share = true;
3209             break;
3210         case OPTION_OUTPUT:
3211             output = optarg;
3212             break;
3213         case 's':
3214             start_offset = cvtnum("start offset", optarg);
3215             if (start_offset < 0) {
3216                 return 1;
3217             }
3218             break;
3219         case 'l':
3220             max_length = cvtnum("max length", optarg);
3221             if (max_length < 0) {
3222                 return 1;
3223             }
3224             break;
3225         case OPTION_OBJECT:
3226             user_creatable_process_cmdline(optarg);
3227             break;
3228         case OPTION_IMAGE_OPTS:
3229             image_opts = true;
3230             break;
3231         }
3232     }
3233     if (optind != argc - 1) {
3234         error_exit("Expecting one image file name");
3235     }
3236     filename = argv[optind];
3237 
3238     if (output && !strcmp(output, "json")) {
3239         output_format = OFORMAT_JSON;
3240     } else if (output && !strcmp(output, "human")) {
3241         output_format = OFORMAT_HUMAN;
3242     } else if (output) {
3243         error_report("--output must be used with human or json as argument.");
3244         return 1;
3245     }
3246 
3247     blk = img_open(image_opts, filename, fmt, 0, false, false, force_share);
3248     if (!blk) {
3249         return 1;
3250     }
3251     bs = blk_bs(blk);
3252 
3253     if (output_format == OFORMAT_HUMAN) {
3254         printf("%-16s%-16s%-16s%s\n", "Offset", "Length", "Mapped to", "File");
3255     } else if (output_format == OFORMAT_JSON) {
3256         putchar('[');
3257     }
3258 
3259     length = blk_getlength(blk);
3260     if (length < 0) {
3261         error_report("Failed to get size for '%s'", filename);
3262         return 1;
3263     }
3264     if (max_length != -1) {
3265         length = MIN(start_offset + max_length, length);
3266     }
3267 
3268     curr.start = start_offset;
3269     while (curr.start + curr.length < length) {
3270         int64_t offset = curr.start + curr.length;
3271         int64_t n = length - offset;
3272 
3273         ret = get_block_status(bs, offset, n, &next);
3274         if (ret < 0) {
3275             error_report("Could not read file metadata: %s", strerror(-ret));
3276             goto out;
3277         }
3278 
3279         if (entry_mergeable(&curr, &next)) {
3280             curr.length += next.length;
3281             continue;
3282         }
3283 
3284         if (curr.length > 0) {
3285             ret = dump_map_entry(output_format, &curr, &next);
3286             if (ret < 0) {
3287                 goto out;
3288             }
3289         }
3290         curr = next;
3291     }
3292 
3293     ret = dump_map_entry(output_format, &curr, NULL);
3294     if (output_format == OFORMAT_JSON) {
3295         puts("]");
3296     }
3297 
3298 out:
3299     blk_unref(blk);
3300     return ret < 0;
3301 }
3302 
3303 #define SNAPSHOT_LIST   1
3304 #define SNAPSHOT_CREATE 2
3305 #define SNAPSHOT_APPLY  3
3306 #define SNAPSHOT_DELETE 4
3307 
3308 static int img_snapshot(int argc, char **argv)
3309 {
3310     BlockBackend *blk;
3311     BlockDriverState *bs;
3312     QEMUSnapshotInfo sn;
3313     char *filename, *snapshot_name = NULL;
3314     int c, ret = 0, bdrv_oflags;
3315     int action = 0;
3316     bool quiet = false;
3317     Error *err = NULL;
3318     bool image_opts = false;
3319     bool force_share = false;
3320     int64_t rt;
3321 
3322     bdrv_oflags = BDRV_O_RDWR;
3323     /* Parse commandline parameters */
3324     for(;;) {
3325         static const struct option long_options[] = {
3326             {"help", no_argument, 0, 'h'},
3327             {"object", required_argument, 0, OPTION_OBJECT},
3328             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3329             {"force-share", no_argument, 0, 'U'},
3330             {0, 0, 0, 0}
3331         };
3332         c = getopt_long(argc, argv, ":la:c:d:hqU",
3333                         long_options, NULL);
3334         if (c == -1) {
3335             break;
3336         }
3337         switch(c) {
3338         case ':':
3339             missing_argument(argv[optind - 1]);
3340             break;
3341         case '?':
3342             unrecognized_option(argv[optind - 1]);
3343             break;
3344         case 'h':
3345             help();
3346             return 0;
3347         case 'l':
3348             if (action) {
3349                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3350                 return 0;
3351             }
3352             action = SNAPSHOT_LIST;
3353             bdrv_oflags &= ~BDRV_O_RDWR; /* no need for RW */
3354             break;
3355         case 'a':
3356             if (action) {
3357                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3358                 return 0;
3359             }
3360             action = SNAPSHOT_APPLY;
3361             snapshot_name = optarg;
3362             break;
3363         case 'c':
3364             if (action) {
3365                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3366                 return 0;
3367             }
3368             action = SNAPSHOT_CREATE;
3369             snapshot_name = optarg;
3370             break;
3371         case 'd':
3372             if (action) {
3373                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3374                 return 0;
3375             }
3376             action = SNAPSHOT_DELETE;
3377             snapshot_name = optarg;
3378             break;
3379         case 'q':
3380             quiet = true;
3381             break;
3382         case 'U':
3383             force_share = true;
3384             break;
3385         case OPTION_OBJECT:
3386             user_creatable_process_cmdline(optarg);
3387             break;
3388         case OPTION_IMAGE_OPTS:
3389             image_opts = true;
3390             break;
3391         }
3392     }
3393 
3394     if (optind != argc - 1) {
3395         error_exit("Expecting one image file name");
3396     }
3397     filename = argv[optind++];
3398 
3399     /* Open the image */
3400     blk = img_open(image_opts, filename, NULL, bdrv_oflags, false, quiet,
3401                    force_share);
3402     if (!blk) {
3403         return 1;
3404     }
3405     bs = blk_bs(blk);
3406 
3407     /* Perform the requested action */
3408     switch(action) {
3409     case SNAPSHOT_LIST:
3410         dump_snapshots(bs);
3411         break;
3412 
3413     case SNAPSHOT_CREATE:
3414         memset(&sn, 0, sizeof(sn));
3415         pstrcpy(sn.name, sizeof(sn.name), snapshot_name);
3416 
3417         rt = g_get_real_time();
3418         sn.date_sec = rt / G_USEC_PER_SEC;
3419         sn.date_nsec = (rt % G_USEC_PER_SEC) * 1000;
3420 
3421         ret = bdrv_snapshot_create(bs, &sn);
3422         if (ret) {
3423             error_report("Could not create snapshot '%s': %d (%s)",
3424                 snapshot_name, ret, strerror(-ret));
3425         }
3426         break;
3427 
3428     case SNAPSHOT_APPLY:
3429         ret = bdrv_snapshot_goto(bs, snapshot_name, &err);
3430         if (ret) {
3431             error_reportf_err(err, "Could not apply snapshot '%s': ",
3432                               snapshot_name);
3433         }
3434         break;
3435 
3436     case SNAPSHOT_DELETE:
3437         ret = bdrv_snapshot_find(bs, &sn, snapshot_name);
3438         if (ret < 0) {
3439             error_report("Could not delete snapshot '%s': snapshot not "
3440                          "found", snapshot_name);
3441             ret = 1;
3442         } else {
3443             ret = bdrv_snapshot_delete(bs, sn.id_str, sn.name, &err);
3444             if (ret < 0) {
3445                 error_reportf_err(err, "Could not delete snapshot '%s': ",
3446                                   snapshot_name);
3447                 ret = 1;
3448             }
3449         }
3450         break;
3451     }
3452 
3453     /* Cleanup */
3454     blk_unref(blk);
3455     if (ret) {
3456         return 1;
3457     }
3458     return 0;
3459 }
3460 
3461 static int img_rebase(int argc, char **argv)
3462 {
3463     BlockBackend *blk = NULL, *blk_old_backing = NULL, *blk_new_backing = NULL;
3464     uint8_t *buf_old = NULL;
3465     uint8_t *buf_new = NULL;
3466     BlockDriverState *bs = NULL, *prefix_chain_bs = NULL;
3467     BlockDriverState *unfiltered_bs;
3468     char *filename;
3469     const char *fmt, *cache, *src_cache, *out_basefmt, *out_baseimg;
3470     int c, flags, src_flags, ret;
3471     bool writethrough, src_writethrough;
3472     int unsafe = 0;
3473     bool force_share = false;
3474     int progress = 0;
3475     bool quiet = false;
3476     Error *local_err = NULL;
3477     bool image_opts = false;
3478 
3479     /* Parse commandline parameters */
3480     fmt = NULL;
3481     cache = BDRV_DEFAULT_CACHE;
3482     src_cache = BDRV_DEFAULT_CACHE;
3483     out_baseimg = NULL;
3484     out_basefmt = NULL;
3485     for(;;) {
3486         static const struct option long_options[] = {
3487             {"help", no_argument, 0, 'h'},
3488             {"object", required_argument, 0, OPTION_OBJECT},
3489             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3490             {"force-share", no_argument, 0, 'U'},
3491             {0, 0, 0, 0}
3492         };
3493         c = getopt_long(argc, argv, ":hf:F:b:upt:T:qU",
3494                         long_options, NULL);
3495         if (c == -1) {
3496             break;
3497         }
3498         switch(c) {
3499         case ':':
3500             missing_argument(argv[optind - 1]);
3501             break;
3502         case '?':
3503             unrecognized_option(argv[optind - 1]);
3504             break;
3505         case 'h':
3506             help();
3507             return 0;
3508         case 'f':
3509             fmt = optarg;
3510             break;
3511         case 'F':
3512             out_basefmt = optarg;
3513             break;
3514         case 'b':
3515             out_baseimg = optarg;
3516             break;
3517         case 'u':
3518             unsafe = 1;
3519             break;
3520         case 'p':
3521             progress = 1;
3522             break;
3523         case 't':
3524             cache = optarg;
3525             break;
3526         case 'T':
3527             src_cache = optarg;
3528             break;
3529         case 'q':
3530             quiet = true;
3531             break;
3532         case OPTION_OBJECT:
3533             user_creatable_process_cmdline(optarg);
3534             break;
3535         case OPTION_IMAGE_OPTS:
3536             image_opts = true;
3537             break;
3538         case 'U':
3539             force_share = true;
3540             break;
3541         }
3542     }
3543 
3544     if (quiet) {
3545         progress = 0;
3546     }
3547 
3548     if (optind != argc - 1) {
3549         error_exit("Expecting one image file name");
3550     }
3551     if (!unsafe && !out_baseimg) {
3552         error_exit("Must specify backing file (-b) or use unsafe mode (-u)");
3553     }
3554     filename = argv[optind++];
3555 
3556     qemu_progress_init(progress, 2.0);
3557     qemu_progress_print(0, 100);
3558 
3559     flags = BDRV_O_RDWR | (unsafe ? BDRV_O_NO_BACKING : 0);
3560     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
3561     if (ret < 0) {
3562         error_report("Invalid cache option: %s", cache);
3563         goto out;
3564     }
3565 
3566     src_flags = 0;
3567     ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
3568     if (ret < 0) {
3569         error_report("Invalid source cache option: %s", src_cache);
3570         goto out;
3571     }
3572 
3573     /* The source files are opened read-only, don't care about WCE */
3574     assert((src_flags & BDRV_O_RDWR) == 0);
3575     (void) src_writethrough;
3576 
3577     /*
3578      * Open the images.
3579      *
3580      * Ignore the old backing file for unsafe rebase in case we want to correct
3581      * the reference to a renamed or moved backing file.
3582      */
3583     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
3584                    false);
3585     if (!blk) {
3586         ret = -1;
3587         goto out;
3588     }
3589     bs = blk_bs(blk);
3590 
3591     unfiltered_bs = bdrv_skip_filters(bs);
3592 
3593     if (out_basefmt != NULL) {
3594         if (bdrv_find_format(out_basefmt) == NULL) {
3595             error_report("Invalid format name: '%s'", out_basefmt);
3596             ret = -1;
3597             goto out;
3598         }
3599     }
3600 
3601     /* For safe rebasing we need to compare old and new backing file */
3602     if (!unsafe) {
3603         QDict *options = NULL;
3604         BlockDriverState *base_bs = bdrv_cow_bs(unfiltered_bs);
3605 
3606         if (base_bs) {
3607             blk_old_backing = blk_new(qemu_get_aio_context(),
3608                                       BLK_PERM_CONSISTENT_READ,
3609                                       BLK_PERM_ALL);
3610             ret = blk_insert_bs(blk_old_backing, base_bs,
3611                                 &local_err);
3612             if (ret < 0) {
3613                 error_reportf_err(local_err,
3614                                   "Could not reuse old backing file '%s': ",
3615                                   base_bs->filename);
3616                 goto out;
3617             }
3618         } else {
3619             blk_old_backing = NULL;
3620         }
3621 
3622         if (out_baseimg[0]) {
3623             const char *overlay_filename;
3624             char *out_real_path;
3625 
3626             options = qdict_new();
3627             if (out_basefmt) {
3628                 qdict_put_str(options, "driver", out_basefmt);
3629             }
3630             if (force_share) {
3631                 qdict_put_bool(options, BDRV_OPT_FORCE_SHARE, true);
3632             }
3633 
3634             bdrv_refresh_filename(bs);
3635             overlay_filename = bs->exact_filename[0] ? bs->exact_filename
3636                                                      : bs->filename;
3637             out_real_path =
3638                 bdrv_get_full_backing_filename_from_filename(overlay_filename,
3639                                                              out_baseimg,
3640                                                              &local_err);
3641             if (local_err) {
3642                 qobject_unref(options);
3643                 error_reportf_err(local_err,
3644                                   "Could not resolve backing filename: ");
3645                 ret = -1;
3646                 goto out;
3647             }
3648 
3649             /*
3650              * Find out whether we rebase an image on top of a previous image
3651              * in its chain.
3652              */
3653             prefix_chain_bs = bdrv_find_backing_image(bs, out_real_path);
3654             if (prefix_chain_bs) {
3655                 qobject_unref(options);
3656                 g_free(out_real_path);
3657 
3658                 blk_new_backing = blk_new(qemu_get_aio_context(),
3659                                           BLK_PERM_CONSISTENT_READ,
3660                                           BLK_PERM_ALL);
3661                 ret = blk_insert_bs(blk_new_backing, prefix_chain_bs,
3662                                     &local_err);
3663                 if (ret < 0) {
3664                     error_reportf_err(local_err,
3665                                       "Could not reuse backing file '%s': ",
3666                                       out_baseimg);
3667                     goto out;
3668                 }
3669             } else {
3670                 blk_new_backing = blk_new_open(out_real_path, NULL,
3671                                                options, src_flags, &local_err);
3672                 g_free(out_real_path);
3673                 if (!blk_new_backing) {
3674                     error_reportf_err(local_err,
3675                                       "Could not open new backing file '%s': ",
3676                                       out_baseimg);
3677                     ret = -1;
3678                     goto out;
3679                 }
3680             }
3681         }
3682     }
3683 
3684     /*
3685      * Check each unallocated cluster in the COW file. If it is unallocated,
3686      * accesses go to the backing file. We must therefore compare this cluster
3687      * in the old and new backing file, and if they differ we need to copy it
3688      * from the old backing file into the COW file.
3689      *
3690      * If qemu-img crashes during this step, no harm is done. The content of
3691      * the image is the same as the original one at any time.
3692      */
3693     if (!unsafe) {
3694         int64_t size;
3695         int64_t old_backing_size = 0;
3696         int64_t new_backing_size = 0;
3697         uint64_t offset;
3698         int64_t n;
3699         float local_progress = 0;
3700 
3701         buf_old = blk_blockalign(blk, IO_BUF_SIZE);
3702         buf_new = blk_blockalign(blk, IO_BUF_SIZE);
3703 
3704         size = blk_getlength(blk);
3705         if (size < 0) {
3706             error_report("Could not get size of '%s': %s",
3707                          filename, strerror(-size));
3708             ret = -1;
3709             goto out;
3710         }
3711         if (blk_old_backing) {
3712             old_backing_size = blk_getlength(blk_old_backing);
3713             if (old_backing_size < 0) {
3714                 char backing_name[PATH_MAX];
3715 
3716                 bdrv_get_backing_filename(bs, backing_name,
3717                                           sizeof(backing_name));
3718                 error_report("Could not get size of '%s': %s",
3719                              backing_name, strerror(-old_backing_size));
3720                 ret = -1;
3721                 goto out;
3722             }
3723         }
3724         if (blk_new_backing) {
3725             new_backing_size = blk_getlength(blk_new_backing);
3726             if (new_backing_size < 0) {
3727                 error_report("Could not get size of '%s': %s",
3728                              out_baseimg, strerror(-new_backing_size));
3729                 ret = -1;
3730                 goto out;
3731             }
3732         }
3733 
3734         if (size != 0) {
3735             local_progress = (float)100 / (size / MIN(size, IO_BUF_SIZE));
3736         }
3737 
3738         for (offset = 0; offset < size; offset += n) {
3739             bool buf_old_is_zero = false;
3740 
3741             /* How many bytes can we handle with the next read? */
3742             n = MIN(IO_BUF_SIZE, size - offset);
3743 
3744             /* If the cluster is allocated, we don't need to take action */
3745             ret = bdrv_is_allocated(unfiltered_bs, offset, n, &n);
3746             if (ret < 0) {
3747                 error_report("error while reading image metadata: %s",
3748                              strerror(-ret));
3749                 goto out;
3750             }
3751             if (ret) {
3752                 continue;
3753             }
3754 
3755             if (prefix_chain_bs) {
3756                 /*
3757                  * If cluster wasn't changed since prefix_chain, we don't need
3758                  * to take action
3759                  */
3760                 ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
3761                                               prefix_chain_bs, false,
3762                                               offset, n, &n);
3763                 if (ret < 0) {
3764                     error_report("error while reading image metadata: %s",
3765                                  strerror(-ret));
3766                     goto out;
3767                 }
3768                 if (!ret) {
3769                     continue;
3770                 }
3771             }
3772 
3773             /*
3774              * Read old and new backing file and take into consideration that
3775              * backing files may be smaller than the COW image.
3776              */
3777             if (offset >= old_backing_size) {
3778                 memset(buf_old, 0, n);
3779                 buf_old_is_zero = true;
3780             } else {
3781                 if (offset + n > old_backing_size) {
3782                     n = old_backing_size - offset;
3783                 }
3784 
3785                 ret = blk_pread(blk_old_backing, offset, n, buf_old, 0);
3786                 if (ret < 0) {
3787                     error_report("error while reading from old backing file");
3788                     goto out;
3789                 }
3790             }
3791 
3792             if (offset >= new_backing_size || !blk_new_backing) {
3793                 memset(buf_new, 0, n);
3794             } else {
3795                 if (offset + n > new_backing_size) {
3796                     n = new_backing_size - offset;
3797                 }
3798 
3799                 ret = blk_pread(blk_new_backing, offset, n, buf_new, 0);
3800                 if (ret < 0) {
3801                     error_report("error while reading from new backing file");
3802                     goto out;
3803                 }
3804             }
3805 
3806             /* If they differ, we need to write to the COW file */
3807             uint64_t written = 0;
3808 
3809             while (written < n) {
3810                 int64_t pnum;
3811 
3812                 if (compare_buffers(buf_old + written, buf_new + written,
3813                                     n - written, &pnum))
3814                 {
3815                     if (buf_old_is_zero) {
3816                         ret = blk_pwrite_zeroes(blk, offset + written, pnum, 0);
3817                     } else {
3818                         ret = blk_pwrite(blk, offset + written, pnum,
3819                                          buf_old + written, 0);
3820                     }
3821                     if (ret < 0) {
3822                         error_report("Error while writing to COW image: %s",
3823                             strerror(-ret));
3824                         goto out;
3825                     }
3826                 }
3827 
3828                 written += pnum;
3829             }
3830             qemu_progress_print(local_progress, 100);
3831         }
3832     }
3833 
3834     /*
3835      * Change the backing file. All clusters that are different from the old
3836      * backing file are overwritten in the COW file now, so the visible content
3837      * doesn't change when we switch the backing file.
3838      */
3839     if (out_baseimg && *out_baseimg) {
3840         ret = bdrv_change_backing_file(unfiltered_bs, out_baseimg, out_basefmt,
3841                                        true);
3842     } else {
3843         ret = bdrv_change_backing_file(unfiltered_bs, NULL, NULL, false);
3844     }
3845 
3846     if (ret == -ENOSPC) {
3847         error_report("Could not change the backing file to '%s': No "
3848                      "space left in the file header", out_baseimg);
3849     } else if (ret == -EINVAL && out_baseimg && !out_basefmt) {
3850         error_report("Could not change the backing file to '%s': backing "
3851                      "format must be specified", out_baseimg);
3852     } else if (ret < 0) {
3853         error_report("Could not change the backing file to '%s': %s",
3854             out_baseimg, strerror(-ret));
3855     }
3856 
3857     qemu_progress_print(100, 0);
3858     /*
3859      * TODO At this point it is possible to check if any clusters that are
3860      * allocated in the COW file are the same in the backing file. If so, they
3861      * could be dropped from the COW file. Don't do this before switching the
3862      * backing file, in case of a crash this would lead to corruption.
3863      */
3864 out:
3865     qemu_progress_end();
3866     /* Cleanup */
3867     if (!unsafe) {
3868         blk_unref(blk_old_backing);
3869         blk_unref(blk_new_backing);
3870     }
3871     qemu_vfree(buf_old);
3872     qemu_vfree(buf_new);
3873 
3874     blk_unref(blk);
3875     if (ret) {
3876         return 1;
3877     }
3878     return 0;
3879 }
3880 
3881 static int img_resize(int argc, char **argv)
3882 {
3883     Error *err = NULL;
3884     int c, ret, relative;
3885     const char *filename, *fmt, *size;
3886     int64_t n, total_size, current_size;
3887     bool quiet = false;
3888     BlockBackend *blk = NULL;
3889     PreallocMode prealloc = PREALLOC_MODE_OFF;
3890     QemuOpts *param;
3891 
3892     static QemuOptsList resize_options = {
3893         .name = "resize_options",
3894         .head = QTAILQ_HEAD_INITIALIZER(resize_options.head),
3895         .desc = {
3896             {
3897                 .name = BLOCK_OPT_SIZE,
3898                 .type = QEMU_OPT_SIZE,
3899                 .help = "Virtual disk size"
3900             }, {
3901                 /* end of list */
3902             }
3903         },
3904     };
3905     bool image_opts = false;
3906     bool shrink = false;
3907 
3908     /* Remove size from argv manually so that negative numbers are not treated
3909      * as options by getopt. */
3910     if (argc < 3) {
3911         error_exit("Not enough arguments");
3912         return 1;
3913     }
3914 
3915     size = argv[--argc];
3916 
3917     /* Parse getopt arguments */
3918     fmt = NULL;
3919     for(;;) {
3920         static const struct option long_options[] = {
3921             {"help", no_argument, 0, 'h'},
3922             {"object", required_argument, 0, OPTION_OBJECT},
3923             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3924             {"preallocation", required_argument, 0, OPTION_PREALLOCATION},
3925             {"shrink", no_argument, 0, OPTION_SHRINK},
3926             {0, 0, 0, 0}
3927         };
3928         c = getopt_long(argc, argv, ":f:hq",
3929                         long_options, NULL);
3930         if (c == -1) {
3931             break;
3932         }
3933         switch(c) {
3934         case ':':
3935             missing_argument(argv[optind - 1]);
3936             break;
3937         case '?':
3938             unrecognized_option(argv[optind - 1]);
3939             break;
3940         case 'h':
3941             help();
3942             break;
3943         case 'f':
3944             fmt = optarg;
3945             break;
3946         case 'q':
3947             quiet = true;
3948             break;
3949         case OPTION_OBJECT:
3950             user_creatable_process_cmdline(optarg);
3951             break;
3952         case OPTION_IMAGE_OPTS:
3953             image_opts = true;
3954             break;
3955         case OPTION_PREALLOCATION:
3956             prealloc = qapi_enum_parse(&PreallocMode_lookup, optarg,
3957                                        PREALLOC_MODE__MAX, NULL);
3958             if (prealloc == PREALLOC_MODE__MAX) {
3959                 error_report("Invalid preallocation mode '%s'", optarg);
3960                 return 1;
3961             }
3962             break;
3963         case OPTION_SHRINK:
3964             shrink = true;
3965             break;
3966         }
3967     }
3968     if (optind != argc - 1) {
3969         error_exit("Expecting image file name and size");
3970     }
3971     filename = argv[optind++];
3972 
3973     /* Choose grow, shrink, or absolute resize mode */
3974     switch (size[0]) {
3975     case '+':
3976         relative = 1;
3977         size++;
3978         break;
3979     case '-':
3980         relative = -1;
3981         size++;
3982         break;
3983     default:
3984         relative = 0;
3985         break;
3986     }
3987 
3988     /* Parse size */
3989     param = qemu_opts_create(&resize_options, NULL, 0, &error_abort);
3990     if (!qemu_opt_set(param, BLOCK_OPT_SIZE, size, &err)) {
3991         error_report_err(err);
3992         ret = -1;
3993         qemu_opts_del(param);
3994         goto out;
3995     }
3996     n = qemu_opt_get_size(param, BLOCK_OPT_SIZE, 0);
3997     qemu_opts_del(param);
3998 
3999     blk = img_open(image_opts, filename, fmt,
4000                    BDRV_O_RDWR | BDRV_O_RESIZE, false, quiet,
4001                    false);
4002     if (!blk) {
4003         ret = -1;
4004         goto out;
4005     }
4006 
4007     current_size = blk_getlength(blk);
4008     if (current_size < 0) {
4009         error_report("Failed to inquire current image length: %s",
4010                      strerror(-current_size));
4011         ret = -1;
4012         goto out;
4013     }
4014 
4015     if (relative) {
4016         total_size = current_size + n * relative;
4017     } else {
4018         total_size = n;
4019     }
4020     if (total_size <= 0) {
4021         error_report("New image size must be positive");
4022         ret = -1;
4023         goto out;
4024     }
4025 
4026     if (total_size <= current_size && prealloc != PREALLOC_MODE_OFF) {
4027         error_report("Preallocation can only be used for growing images");
4028         ret = -1;
4029         goto out;
4030     }
4031 
4032     if (total_size < current_size && !shrink) {
4033         error_report("Use the --shrink option to perform a shrink operation.");
4034         warn_report("Shrinking an image will delete all data beyond the "
4035                     "shrunken image's end. Before performing such an "
4036                     "operation, make sure there is no important data there.");
4037         ret = -1;
4038         goto out;
4039     }
4040 
4041     /*
4042      * The user expects the image to have the desired size after
4043      * resizing, so pass @exact=true.  It is of no use to report
4044      * success when the image has not actually been resized.
4045      */
4046     ret = blk_truncate(blk, total_size, true, prealloc, 0, &err);
4047     if (!ret) {
4048         qprintf(quiet, "Image resized.\n");
4049     } else {
4050         error_report_err(err);
4051     }
4052 out:
4053     blk_unref(blk);
4054     if (ret) {
4055         return 1;
4056     }
4057     return 0;
4058 }
4059 
4060 static void amend_status_cb(BlockDriverState *bs,
4061                             int64_t offset, int64_t total_work_size,
4062                             void *opaque)
4063 {
4064     qemu_progress_print(100.f * offset / total_work_size, 0);
4065 }
4066 
4067 static int print_amend_option_help(const char *format)
4068 {
4069     BlockDriver *drv;
4070 
4071     /* Find driver and parse its options */
4072     drv = bdrv_find_format(format);
4073     if (!drv) {
4074         error_report("Unknown file format '%s'", format);
4075         return 1;
4076     }
4077 
4078     if (!drv->bdrv_amend_options) {
4079         error_report("Format driver '%s' does not support option amendment",
4080                      format);
4081         return 1;
4082     }
4083 
4084     /* Every driver supporting amendment must have amend_opts */
4085     assert(drv->amend_opts);
4086 
4087     printf("Amend options for '%s':\n", format);
4088     qemu_opts_print_help(drv->amend_opts, false);
4089     return 0;
4090 }
4091 
4092 static int img_amend(int argc, char **argv)
4093 {
4094     Error *err = NULL;
4095     int c, ret = 0;
4096     char *options = NULL;
4097     QemuOptsList *amend_opts = NULL;
4098     QemuOpts *opts = NULL;
4099     const char *fmt = NULL, *filename, *cache;
4100     int flags;
4101     bool writethrough;
4102     bool quiet = false, progress = false;
4103     BlockBackend *blk = NULL;
4104     BlockDriverState *bs = NULL;
4105     bool image_opts = false;
4106     bool force = false;
4107 
4108     cache = BDRV_DEFAULT_CACHE;
4109     for (;;) {
4110         static const struct option long_options[] = {
4111             {"help", no_argument, 0, 'h'},
4112             {"object", required_argument, 0, OPTION_OBJECT},
4113             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4114             {"force", no_argument, 0, OPTION_FORCE},
4115             {0, 0, 0, 0}
4116         };
4117         c = getopt_long(argc, argv, ":ho:f:t:pq",
4118                         long_options, NULL);
4119         if (c == -1) {
4120             break;
4121         }
4122 
4123         switch (c) {
4124         case ':':
4125             missing_argument(argv[optind - 1]);
4126             break;
4127         case '?':
4128             unrecognized_option(argv[optind - 1]);
4129             break;
4130         case 'h':
4131             help();
4132             break;
4133         case 'o':
4134             if (accumulate_options(&options, optarg) < 0) {
4135                 ret = -1;
4136                 goto out_no_progress;
4137             }
4138             break;
4139         case 'f':
4140             fmt = optarg;
4141             break;
4142         case 't':
4143             cache = optarg;
4144             break;
4145         case 'p':
4146             progress = true;
4147             break;
4148         case 'q':
4149             quiet = true;
4150             break;
4151         case OPTION_OBJECT:
4152             user_creatable_process_cmdline(optarg);
4153             break;
4154         case OPTION_IMAGE_OPTS:
4155             image_opts = true;
4156             break;
4157         case OPTION_FORCE:
4158             force = true;
4159             break;
4160         }
4161     }
4162 
4163     if (!options) {
4164         error_exit("Must specify options (-o)");
4165     }
4166 
4167     if (quiet) {
4168         progress = false;
4169     }
4170     qemu_progress_init(progress, 1.0);
4171 
4172     filename = (optind == argc - 1) ? argv[argc - 1] : NULL;
4173     if (fmt && has_help_option(options)) {
4174         /* If a format is explicitly specified (and possibly no filename is
4175          * given), print option help here */
4176         ret = print_amend_option_help(fmt);
4177         goto out;
4178     }
4179 
4180     if (optind != argc - 1) {
4181         error_report("Expecting one image file name");
4182         ret = -1;
4183         goto out;
4184     }
4185 
4186     flags = BDRV_O_RDWR;
4187     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
4188     if (ret < 0) {
4189         error_report("Invalid cache option: %s", cache);
4190         goto out;
4191     }
4192 
4193     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
4194                    false);
4195     if (!blk) {
4196         ret = -1;
4197         goto out;
4198     }
4199     bs = blk_bs(blk);
4200 
4201     fmt = bs->drv->format_name;
4202 
4203     if (has_help_option(options)) {
4204         /* If the format was auto-detected, print option help here */
4205         ret = print_amend_option_help(fmt);
4206         goto out;
4207     }
4208 
4209     if (!bs->drv->bdrv_amend_options) {
4210         error_report("Format driver '%s' does not support option amendment",
4211                      fmt);
4212         ret = -1;
4213         goto out;
4214     }
4215 
4216     /* Every driver supporting amendment must have amend_opts */
4217     assert(bs->drv->amend_opts);
4218 
4219     amend_opts = qemu_opts_append(amend_opts, bs->drv->amend_opts);
4220     opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
4221     if (!qemu_opts_do_parse(opts, options, NULL, &err)) {
4222         /* Try to parse options using the create options */
4223         amend_opts = qemu_opts_append(amend_opts, bs->drv->create_opts);
4224         qemu_opts_del(opts);
4225         opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
4226         if (qemu_opts_do_parse(opts, options, NULL, NULL)) {
4227             error_append_hint(&err,
4228                               "This option is only supported for image creation\n");
4229         }
4230 
4231         error_report_err(err);
4232         ret = -1;
4233         goto out;
4234     }
4235 
4236     /* In case the driver does not call amend_status_cb() */
4237     qemu_progress_print(0.f, 0);
4238     ret = bdrv_amend_options(bs, opts, &amend_status_cb, NULL, force, &err);
4239     qemu_progress_print(100.f, 0);
4240     if (ret < 0) {
4241         error_report_err(err);
4242         goto out;
4243     }
4244 
4245 out:
4246     qemu_progress_end();
4247 
4248 out_no_progress:
4249     blk_unref(blk);
4250     qemu_opts_del(opts);
4251     qemu_opts_free(amend_opts);
4252     g_free(options);
4253 
4254     if (ret) {
4255         return 1;
4256     }
4257     return 0;
4258 }
4259 
4260 typedef struct BenchData {
4261     BlockBackend *blk;
4262     uint64_t image_size;
4263     bool write;
4264     int bufsize;
4265     int step;
4266     int nrreq;
4267     int n;
4268     int flush_interval;
4269     bool drain_on_flush;
4270     uint8_t *buf;
4271     QEMUIOVector *qiov;
4272 
4273     int in_flight;
4274     bool in_flush;
4275     uint64_t offset;
4276 } BenchData;
4277 
4278 static void bench_undrained_flush_cb(void *opaque, int ret)
4279 {
4280     if (ret < 0) {
4281         error_report("Failed flush request: %s", strerror(-ret));
4282         exit(EXIT_FAILURE);
4283     }
4284 }
4285 
4286 static void bench_cb(void *opaque, int ret)
4287 {
4288     BenchData *b = opaque;
4289     BlockAIOCB *acb;
4290 
4291     if (ret < 0) {
4292         error_report("Failed request: %s", strerror(-ret));
4293         exit(EXIT_FAILURE);
4294     }
4295 
4296     if (b->in_flush) {
4297         /* Just finished a flush with drained queue: Start next requests */
4298         assert(b->in_flight == 0);
4299         b->in_flush = false;
4300     } else if (b->in_flight > 0) {
4301         int remaining = b->n - b->in_flight;
4302 
4303         b->n--;
4304         b->in_flight--;
4305 
4306         /* Time for flush? Drain queue if requested, then flush */
4307         if (b->flush_interval && remaining % b->flush_interval == 0) {
4308             if (!b->in_flight || !b->drain_on_flush) {
4309                 BlockCompletionFunc *cb;
4310 
4311                 if (b->drain_on_flush) {
4312                     b->in_flush = true;
4313                     cb = bench_cb;
4314                 } else {
4315                     cb = bench_undrained_flush_cb;
4316                 }
4317 
4318                 acb = blk_aio_flush(b->blk, cb, b);
4319                 if (!acb) {
4320                     error_report("Failed to issue flush request");
4321                     exit(EXIT_FAILURE);
4322                 }
4323             }
4324             if (b->drain_on_flush) {
4325                 return;
4326             }
4327         }
4328     }
4329 
4330     while (b->n > b->in_flight && b->in_flight < b->nrreq) {
4331         int64_t offset = b->offset;
4332         /* blk_aio_* might look for completed I/Os and kick bench_cb
4333          * again, so make sure this operation is counted by in_flight
4334          * and b->offset is ready for the next submission.
4335          */
4336         b->in_flight++;
4337         b->offset += b->step;
4338         b->offset %= b->image_size;
4339         if (b->write) {
4340             acb = blk_aio_pwritev(b->blk, offset, b->qiov, 0, bench_cb, b);
4341         } else {
4342             acb = blk_aio_preadv(b->blk, offset, b->qiov, 0, bench_cb, b);
4343         }
4344         if (!acb) {
4345             error_report("Failed to issue request");
4346             exit(EXIT_FAILURE);
4347         }
4348     }
4349 }
4350 
4351 static int img_bench(int argc, char **argv)
4352 {
4353     int c, ret = 0;
4354     const char *fmt = NULL, *filename;
4355     bool quiet = false;
4356     bool image_opts = false;
4357     bool is_write = false;
4358     int count = 75000;
4359     int depth = 64;
4360     int64_t offset = 0;
4361     size_t bufsize = 4096;
4362     int pattern = 0;
4363     size_t step = 0;
4364     int flush_interval = 0;
4365     bool drain_on_flush = true;
4366     int64_t image_size;
4367     BlockBackend *blk = NULL;
4368     BenchData data = {};
4369     int flags = 0;
4370     bool writethrough = false;
4371     struct timeval t1, t2;
4372     int i;
4373     bool force_share = false;
4374     size_t buf_size = 0;
4375 
4376     for (;;) {
4377         static const struct option long_options[] = {
4378             {"help", no_argument, 0, 'h'},
4379             {"flush-interval", required_argument, 0, OPTION_FLUSH_INTERVAL},
4380             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4381             {"pattern", required_argument, 0, OPTION_PATTERN},
4382             {"no-drain", no_argument, 0, OPTION_NO_DRAIN},
4383             {"force-share", no_argument, 0, 'U'},
4384             {0, 0, 0, 0}
4385         };
4386         c = getopt_long(argc, argv, ":hc:d:f:ni:o:qs:S:t:wU", long_options,
4387                         NULL);
4388         if (c == -1) {
4389             break;
4390         }
4391 
4392         switch (c) {
4393         case ':':
4394             missing_argument(argv[optind - 1]);
4395             break;
4396         case '?':
4397             unrecognized_option(argv[optind - 1]);
4398             break;
4399         case 'h':
4400             help();
4401             break;
4402         case 'c':
4403         {
4404             unsigned long res;
4405 
4406             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
4407                 error_report("Invalid request count specified");
4408                 return 1;
4409             }
4410             count = res;
4411             break;
4412         }
4413         case 'd':
4414         {
4415             unsigned long res;
4416 
4417             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
4418                 error_report("Invalid queue depth specified");
4419                 return 1;
4420             }
4421             depth = res;
4422             break;
4423         }
4424         case 'f':
4425             fmt = optarg;
4426             break;
4427         case 'n':
4428             flags |= BDRV_O_NATIVE_AIO;
4429             break;
4430         case 'i':
4431             ret = bdrv_parse_aio(optarg, &flags);
4432             if (ret < 0) {
4433                 error_report("Invalid aio option: %s", optarg);
4434                 ret = -1;
4435                 goto out;
4436             }
4437             break;
4438         case 'o':
4439         {
4440             offset = cvtnum("offset", optarg);
4441             if (offset < 0) {
4442                 return 1;
4443             }
4444             break;
4445         }
4446             break;
4447         case 'q':
4448             quiet = true;
4449             break;
4450         case 's':
4451         {
4452             int64_t sval;
4453 
4454             sval = cvtnum_full("buffer size", optarg, 0, INT_MAX);
4455             if (sval < 0) {
4456                 return 1;
4457             }
4458 
4459             bufsize = sval;
4460             break;
4461         }
4462         case 'S':
4463         {
4464             int64_t sval;
4465 
4466             sval = cvtnum_full("step_size", optarg, 0, INT_MAX);
4467             if (sval < 0) {
4468                 return 1;
4469             }
4470 
4471             step = sval;
4472             break;
4473         }
4474         case 't':
4475             ret = bdrv_parse_cache_mode(optarg, &flags, &writethrough);
4476             if (ret < 0) {
4477                 error_report("Invalid cache mode");
4478                 ret = -1;
4479                 goto out;
4480             }
4481             break;
4482         case 'w':
4483             flags |= BDRV_O_RDWR;
4484             is_write = true;
4485             break;
4486         case 'U':
4487             force_share = true;
4488             break;
4489         case OPTION_PATTERN:
4490         {
4491             unsigned long res;
4492 
4493             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > 0xff) {
4494                 error_report("Invalid pattern byte specified");
4495                 return 1;
4496             }
4497             pattern = res;
4498             break;
4499         }
4500         case OPTION_FLUSH_INTERVAL:
4501         {
4502             unsigned long res;
4503 
4504             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
4505                 error_report("Invalid flush interval specified");
4506                 return 1;
4507             }
4508             flush_interval = res;
4509             break;
4510         }
4511         case OPTION_NO_DRAIN:
4512             drain_on_flush = false;
4513             break;
4514         case OPTION_IMAGE_OPTS:
4515             image_opts = true;
4516             break;
4517         }
4518     }
4519 
4520     if (optind != argc - 1) {
4521         error_exit("Expecting one image file name");
4522     }
4523     filename = argv[argc - 1];
4524 
4525     if (!is_write && flush_interval) {
4526         error_report("--flush-interval is only available in write tests");
4527         ret = -1;
4528         goto out;
4529     }
4530     if (flush_interval && flush_interval < depth) {
4531         error_report("Flush interval can't be smaller than depth");
4532         ret = -1;
4533         goto out;
4534     }
4535 
4536     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
4537                    force_share);
4538     if (!blk) {
4539         ret = -1;
4540         goto out;
4541     }
4542 
4543     image_size = blk_getlength(blk);
4544     if (image_size < 0) {
4545         ret = image_size;
4546         goto out;
4547     }
4548 
4549     data = (BenchData) {
4550         .blk            = blk,
4551         .image_size     = image_size,
4552         .bufsize        = bufsize,
4553         .step           = step ?: bufsize,
4554         .nrreq          = depth,
4555         .n              = count,
4556         .offset         = offset,
4557         .write          = is_write,
4558         .flush_interval = flush_interval,
4559         .drain_on_flush = drain_on_flush,
4560     };
4561     printf("Sending %d %s requests, %d bytes each, %d in parallel "
4562            "(starting at offset %" PRId64 ", step size %d)\n",
4563            data.n, data.write ? "write" : "read", data.bufsize, data.nrreq,
4564            data.offset, data.step);
4565     if (flush_interval) {
4566         printf("Sending flush every %d requests\n", flush_interval);
4567     }
4568 
4569     buf_size = data.nrreq * data.bufsize;
4570     data.buf = blk_blockalign(blk, buf_size);
4571     memset(data.buf, pattern, data.nrreq * data.bufsize);
4572 
4573     blk_register_buf(blk, data.buf, buf_size, &error_fatal);
4574 
4575     data.qiov = g_new(QEMUIOVector, data.nrreq);
4576     for (i = 0; i < data.nrreq; i++) {
4577         qemu_iovec_init(&data.qiov[i], 1);
4578         qemu_iovec_add(&data.qiov[i],
4579                        data.buf + i * data.bufsize, data.bufsize);
4580     }
4581 
4582     gettimeofday(&t1, NULL);
4583     bench_cb(&data, 0);
4584 
4585     while (data.n > 0) {
4586         main_loop_wait(false);
4587     }
4588     gettimeofday(&t2, NULL);
4589 
4590     printf("Run completed in %3.3f seconds.\n",
4591            (t2.tv_sec - t1.tv_sec)
4592            + ((double)(t2.tv_usec - t1.tv_usec) / 1000000));
4593 
4594 out:
4595     if (data.buf) {
4596         blk_unregister_buf(blk, data.buf, buf_size);
4597     }
4598     qemu_vfree(data.buf);
4599     blk_unref(blk);
4600 
4601     if (ret) {
4602         return 1;
4603     }
4604     return 0;
4605 }
4606 
4607 enum ImgBitmapAct {
4608     BITMAP_ADD,
4609     BITMAP_REMOVE,
4610     BITMAP_CLEAR,
4611     BITMAP_ENABLE,
4612     BITMAP_DISABLE,
4613     BITMAP_MERGE,
4614 };
4615 typedef struct ImgBitmapAction {
4616     enum ImgBitmapAct act;
4617     const char *src; /* only used for merge */
4618     QSIMPLEQ_ENTRY(ImgBitmapAction) next;
4619 } ImgBitmapAction;
4620 
4621 static int img_bitmap(int argc, char **argv)
4622 {
4623     Error *err = NULL;
4624     int c, ret = 1;
4625     QemuOpts *opts = NULL;
4626     const char *fmt = NULL, *src_fmt = NULL, *src_filename = NULL;
4627     const char *filename, *bitmap;
4628     BlockBackend *blk = NULL, *src = NULL;
4629     BlockDriverState *bs = NULL, *src_bs = NULL;
4630     bool image_opts = false;
4631     int64_t granularity = 0;
4632     bool add = false, merge = false;
4633     QSIMPLEQ_HEAD(, ImgBitmapAction) actions;
4634     ImgBitmapAction *act, *act_next;
4635     const char *op;
4636 
4637     QSIMPLEQ_INIT(&actions);
4638 
4639     for (;;) {
4640         static const struct option long_options[] = {
4641             {"help", no_argument, 0, 'h'},
4642             {"object", required_argument, 0, OPTION_OBJECT},
4643             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4644             {"add", no_argument, 0, OPTION_ADD},
4645             {"remove", no_argument, 0, OPTION_REMOVE},
4646             {"clear", no_argument, 0, OPTION_CLEAR},
4647             {"enable", no_argument, 0, OPTION_ENABLE},
4648             {"disable", no_argument, 0, OPTION_DISABLE},
4649             {"merge", required_argument, 0, OPTION_MERGE},
4650             {"granularity", required_argument, 0, 'g'},
4651             {"source-file", required_argument, 0, 'b'},
4652             {"source-format", required_argument, 0, 'F'},
4653             {0, 0, 0, 0}
4654         };
4655         c = getopt_long(argc, argv, ":b:f:F:g:h", long_options, NULL);
4656         if (c == -1) {
4657             break;
4658         }
4659 
4660         switch (c) {
4661         case ':':
4662             missing_argument(argv[optind - 1]);
4663             break;
4664         case '?':
4665             unrecognized_option(argv[optind - 1]);
4666             break;
4667         case 'h':
4668             help();
4669             break;
4670         case 'b':
4671             src_filename = optarg;
4672             break;
4673         case 'f':
4674             fmt = optarg;
4675             break;
4676         case 'F':
4677             src_fmt = optarg;
4678             break;
4679         case 'g':
4680             granularity = cvtnum("granularity", optarg);
4681             if (granularity < 0) {
4682                 return 1;
4683             }
4684             break;
4685         case OPTION_ADD:
4686             act = g_new0(ImgBitmapAction, 1);
4687             act->act = BITMAP_ADD;
4688             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4689             add = true;
4690             break;
4691         case OPTION_REMOVE:
4692             act = g_new0(ImgBitmapAction, 1);
4693             act->act = BITMAP_REMOVE;
4694             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4695             break;
4696         case OPTION_CLEAR:
4697             act = g_new0(ImgBitmapAction, 1);
4698             act->act = BITMAP_CLEAR;
4699             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4700             break;
4701         case OPTION_ENABLE:
4702             act = g_new0(ImgBitmapAction, 1);
4703             act->act = BITMAP_ENABLE;
4704             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4705             break;
4706         case OPTION_DISABLE:
4707             act = g_new0(ImgBitmapAction, 1);
4708             act->act = BITMAP_DISABLE;
4709             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4710             break;
4711         case OPTION_MERGE:
4712             act = g_new0(ImgBitmapAction, 1);
4713             act->act = BITMAP_MERGE;
4714             act->src = optarg;
4715             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4716             merge = true;
4717             break;
4718         case OPTION_OBJECT:
4719             user_creatable_process_cmdline(optarg);
4720             break;
4721         case OPTION_IMAGE_OPTS:
4722             image_opts = true;
4723             break;
4724         }
4725     }
4726 
4727     if (QSIMPLEQ_EMPTY(&actions)) {
4728         error_report("Need at least one of --add, --remove, --clear, "
4729                      "--enable, --disable, or --merge");
4730         goto out;
4731     }
4732 
4733     if (granularity && !add) {
4734         error_report("granularity only supported with --add");
4735         goto out;
4736     }
4737     if (src_fmt && !src_filename) {
4738         error_report("-F only supported with -b");
4739         goto out;
4740     }
4741     if (src_filename && !merge) {
4742         error_report("Merge bitmap source file only supported with "
4743                      "--merge");
4744         goto out;
4745     }
4746 
4747     if (optind != argc - 2) {
4748         error_report("Expecting filename and bitmap name");
4749         goto out;
4750     }
4751 
4752     filename = argv[optind];
4753     bitmap = argv[optind + 1];
4754 
4755     /*
4756      * No need to open backing chains; we will be manipulating bitmaps
4757      * directly in this image without reference to image contents.
4758      */
4759     blk = img_open(image_opts, filename, fmt, BDRV_O_RDWR | BDRV_O_NO_BACKING,
4760                    false, false, false);
4761     if (!blk) {
4762         goto out;
4763     }
4764     bs = blk_bs(blk);
4765     if (src_filename) {
4766         src = img_open(false, src_filename, src_fmt, BDRV_O_NO_BACKING,
4767                        false, false, false);
4768         if (!src) {
4769             goto out;
4770         }
4771         src_bs = blk_bs(src);
4772     } else {
4773         src_bs = bs;
4774     }
4775 
4776     QSIMPLEQ_FOREACH_SAFE(act, &actions, next, act_next) {
4777         switch (act->act) {
4778         case BITMAP_ADD:
4779             qmp_block_dirty_bitmap_add(bs->node_name, bitmap,
4780                                        !!granularity, granularity, true, true,
4781                                        false, false, &err);
4782             op = "add";
4783             break;
4784         case BITMAP_REMOVE:
4785             qmp_block_dirty_bitmap_remove(bs->node_name, bitmap, &err);
4786             op = "remove";
4787             break;
4788         case BITMAP_CLEAR:
4789             qmp_block_dirty_bitmap_clear(bs->node_name, bitmap, &err);
4790             op = "clear";
4791             break;
4792         case BITMAP_ENABLE:
4793             qmp_block_dirty_bitmap_enable(bs->node_name, bitmap, &err);
4794             op = "enable";
4795             break;
4796         case BITMAP_DISABLE:
4797             qmp_block_dirty_bitmap_disable(bs->node_name, bitmap, &err);
4798             op = "disable";
4799             break;
4800         case BITMAP_MERGE:
4801             do_dirty_bitmap_merge(bs->node_name, bitmap, src_bs->node_name,
4802                                   act->src, &err);
4803             op = "merge";
4804             break;
4805         default:
4806             g_assert_not_reached();
4807         }
4808 
4809         if (err) {
4810             error_reportf_err(err, "Operation %s on bitmap %s failed: ",
4811                               op, bitmap);
4812             goto out;
4813         }
4814         g_free(act);
4815     }
4816 
4817     ret = 0;
4818 
4819  out:
4820     blk_unref(src);
4821     blk_unref(blk);
4822     qemu_opts_del(opts);
4823     return ret;
4824 }
4825 
4826 #define C_BS      01
4827 #define C_COUNT   02
4828 #define C_IF      04
4829 #define C_OF      010
4830 #define C_SKIP    020
4831 
4832 struct DdInfo {
4833     unsigned int flags;
4834     int64_t count;
4835 };
4836 
4837 struct DdIo {
4838     int bsz;    /* Block size */
4839     char *filename;
4840     uint8_t *buf;
4841     int64_t offset;
4842 };
4843 
4844 struct DdOpts {
4845     const char *name;
4846     int (*f)(const char *, struct DdIo *, struct DdIo *, struct DdInfo *);
4847     unsigned int flag;
4848 };
4849 
4850 static int img_dd_bs(const char *arg,
4851                      struct DdIo *in, struct DdIo *out,
4852                      struct DdInfo *dd)
4853 {
4854     int64_t res;
4855 
4856     res = cvtnum_full("bs", arg, 1, INT_MAX);
4857 
4858     if (res < 0) {
4859         return 1;
4860     }
4861     in->bsz = out->bsz = res;
4862 
4863     return 0;
4864 }
4865 
4866 static int img_dd_count(const char *arg,
4867                         struct DdIo *in, struct DdIo *out,
4868                         struct DdInfo *dd)
4869 {
4870     dd->count = cvtnum("count", arg);
4871 
4872     if (dd->count < 0) {
4873         return 1;
4874     }
4875 
4876     return 0;
4877 }
4878 
4879 static int img_dd_if(const char *arg,
4880                      struct DdIo *in, struct DdIo *out,
4881                      struct DdInfo *dd)
4882 {
4883     in->filename = g_strdup(arg);
4884 
4885     return 0;
4886 }
4887 
4888 static int img_dd_of(const char *arg,
4889                      struct DdIo *in, struct DdIo *out,
4890                      struct DdInfo *dd)
4891 {
4892     out->filename = g_strdup(arg);
4893 
4894     return 0;
4895 }
4896 
4897 static int img_dd_skip(const char *arg,
4898                        struct DdIo *in, struct DdIo *out,
4899                        struct DdInfo *dd)
4900 {
4901     in->offset = cvtnum("skip", arg);
4902 
4903     if (in->offset < 0) {
4904         return 1;
4905     }
4906 
4907     return 0;
4908 }
4909 
4910 static int img_dd(int argc, char **argv)
4911 {
4912     int ret = 0;
4913     char *arg = NULL;
4914     char *tmp;
4915     BlockDriver *drv = NULL, *proto_drv = NULL;
4916     BlockBackend *blk1 = NULL, *blk2 = NULL;
4917     QemuOpts *opts = NULL;
4918     QemuOptsList *create_opts = NULL;
4919     Error *local_err = NULL;
4920     bool image_opts = false;
4921     int c, i;
4922     const char *out_fmt = "raw";
4923     const char *fmt = NULL;
4924     int64_t size = 0;
4925     int64_t out_pos, in_pos;
4926     bool force_share = false;
4927     struct DdInfo dd = {
4928         .flags = 0,
4929         .count = 0,
4930     };
4931     struct DdIo in = {
4932         .bsz = 512, /* Block size is by default 512 bytes */
4933         .filename = NULL,
4934         .buf = NULL,
4935         .offset = 0
4936     };
4937     struct DdIo out = {
4938         .bsz = 512,
4939         .filename = NULL,
4940         .buf = NULL,
4941         .offset = 0
4942     };
4943 
4944     const struct DdOpts options[] = {
4945         { "bs", img_dd_bs, C_BS },
4946         { "count", img_dd_count, C_COUNT },
4947         { "if", img_dd_if, C_IF },
4948         { "of", img_dd_of, C_OF },
4949         { "skip", img_dd_skip, C_SKIP },
4950         { NULL, NULL, 0 }
4951     };
4952     const struct option long_options[] = {
4953         { "help", no_argument, 0, 'h'},
4954         { "object", required_argument, 0, OPTION_OBJECT},
4955         { "image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4956         { "force-share", no_argument, 0, 'U'},
4957         { 0, 0, 0, 0 }
4958     };
4959 
4960     while ((c = getopt_long(argc, argv, ":hf:O:U", long_options, NULL))) {
4961         if (c == EOF) {
4962             break;
4963         }
4964         switch (c) {
4965         case 'O':
4966             out_fmt = optarg;
4967             break;
4968         case 'f':
4969             fmt = optarg;
4970             break;
4971         case ':':
4972             missing_argument(argv[optind - 1]);
4973             break;
4974         case '?':
4975             unrecognized_option(argv[optind - 1]);
4976             break;
4977         case 'h':
4978             help();
4979             break;
4980         case 'U':
4981             force_share = true;
4982             break;
4983         case OPTION_OBJECT:
4984             user_creatable_process_cmdline(optarg);
4985             break;
4986         case OPTION_IMAGE_OPTS:
4987             image_opts = true;
4988             break;
4989         }
4990     }
4991 
4992     for (i = optind; i < argc; i++) {
4993         int j;
4994         arg = g_strdup(argv[i]);
4995 
4996         tmp = strchr(arg, '=');
4997         if (tmp == NULL) {
4998             error_report("unrecognized operand %s", arg);
4999             ret = -1;
5000             goto out;
5001         }
5002 
5003         *tmp++ = '\0';
5004 
5005         for (j = 0; options[j].name != NULL; j++) {
5006             if (!strcmp(arg, options[j].name)) {
5007                 break;
5008             }
5009         }
5010         if (options[j].name == NULL) {
5011             error_report("unrecognized operand %s", arg);
5012             ret = -1;
5013             goto out;
5014         }
5015 
5016         if (options[j].f(tmp, &in, &out, &dd) != 0) {
5017             ret = -1;
5018             goto out;
5019         }
5020         dd.flags |= options[j].flag;
5021         g_free(arg);
5022         arg = NULL;
5023     }
5024 
5025     if (!(dd.flags & C_IF && dd.flags & C_OF)) {
5026         error_report("Must specify both input and output files");
5027         ret = -1;
5028         goto out;
5029     }
5030 
5031     blk1 = img_open(image_opts, in.filename, fmt, 0, false, false,
5032                     force_share);
5033 
5034     if (!blk1) {
5035         ret = -1;
5036         goto out;
5037     }
5038 
5039     drv = bdrv_find_format(out_fmt);
5040     if (!drv) {
5041         error_report("Unknown file format");
5042         ret = -1;
5043         goto out;
5044     }
5045     proto_drv = bdrv_find_protocol(out.filename, true, &local_err);
5046 
5047     if (!proto_drv) {
5048         error_report_err(local_err);
5049         ret = -1;
5050         goto out;
5051     }
5052     if (!drv->create_opts) {
5053         error_report("Format driver '%s' does not support image creation",
5054                      drv->format_name);
5055         ret = -1;
5056         goto out;
5057     }
5058     if (!proto_drv->create_opts) {
5059         error_report("Protocol driver '%s' does not support image creation",
5060                      proto_drv->format_name);
5061         ret = -1;
5062         goto out;
5063     }
5064     create_opts = qemu_opts_append(create_opts, drv->create_opts);
5065     create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5066 
5067     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5068 
5069     size = blk_getlength(blk1);
5070     if (size < 0) {
5071         error_report("Failed to get size for '%s'", in.filename);
5072         ret = -1;
5073         goto out;
5074     }
5075 
5076     if (dd.flags & C_COUNT && dd.count <= INT64_MAX / in.bsz &&
5077         dd.count * in.bsz < size) {
5078         size = dd.count * in.bsz;
5079     }
5080 
5081     /* Overflow means the specified offset is beyond input image's size */
5082     if (dd.flags & C_SKIP && (in.offset > INT64_MAX / in.bsz ||
5083                               size < in.bsz * in.offset)) {
5084         qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 0, &error_abort);
5085     } else {
5086         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
5087                             size - in.bsz * in.offset, &error_abort);
5088     }
5089 
5090     ret = bdrv_create(drv, out.filename, opts, &local_err);
5091     if (ret < 0) {
5092         error_reportf_err(local_err,
5093                           "%s: error while creating output image: ",
5094                           out.filename);
5095         ret = -1;
5096         goto out;
5097     }
5098 
5099     /* TODO, we can't honour --image-opts for the target,
5100      * since it needs to be given in a format compatible
5101      * with the bdrv_create() call above which does not
5102      * support image-opts style.
5103      */
5104     blk2 = img_open_file(out.filename, NULL, out_fmt, BDRV_O_RDWR,
5105                          false, false, false);
5106 
5107     if (!blk2) {
5108         ret = -1;
5109         goto out;
5110     }
5111 
5112     if (dd.flags & C_SKIP && (in.offset > INT64_MAX / in.bsz ||
5113                               size < in.offset * in.bsz)) {
5114         /* We give a warning if the skip option is bigger than the input
5115          * size and create an empty output disk image (i.e. like dd(1)).
5116          */
5117         error_report("%s: cannot skip to specified offset", in.filename);
5118         in_pos = size;
5119     } else {
5120         in_pos = in.offset * in.bsz;
5121     }
5122 
5123     in.buf = g_new(uint8_t, in.bsz);
5124 
5125     for (out_pos = 0; in_pos < size; ) {
5126         int bytes = (in_pos + in.bsz > size) ? size - in_pos : in.bsz;
5127 
5128         ret = blk_pread(blk1, in_pos, bytes, in.buf, 0);
5129         if (ret < 0) {
5130             error_report("error while reading from input image file: %s",
5131                          strerror(-ret));
5132             goto out;
5133         }
5134         in_pos += bytes;
5135 
5136         ret = blk_pwrite(blk2, out_pos, bytes, in.buf, 0);
5137         if (ret < 0) {
5138             error_report("error while writing to output image file: %s",
5139                          strerror(-ret));
5140             goto out;
5141         }
5142         out_pos += bytes;
5143     }
5144 
5145 out:
5146     g_free(arg);
5147     qemu_opts_del(opts);
5148     qemu_opts_free(create_opts);
5149     blk_unref(blk1);
5150     blk_unref(blk2);
5151     g_free(in.filename);
5152     g_free(out.filename);
5153     g_free(in.buf);
5154     g_free(out.buf);
5155 
5156     if (ret) {
5157         return 1;
5158     }
5159     return 0;
5160 }
5161 
5162 static void dump_json_block_measure_info(BlockMeasureInfo *info)
5163 {
5164     GString *str;
5165     QObject *obj;
5166     Visitor *v = qobject_output_visitor_new(&obj);
5167 
5168     visit_type_BlockMeasureInfo(v, NULL, &info, &error_abort);
5169     visit_complete(v, &obj);
5170     str = qobject_to_json_pretty(obj, true);
5171     assert(str != NULL);
5172     printf("%s\n", str->str);
5173     qobject_unref(obj);
5174     visit_free(v);
5175     g_string_free(str, true);
5176 }
5177 
5178 static int img_measure(int argc, char **argv)
5179 {
5180     static const struct option long_options[] = {
5181         {"help", no_argument, 0, 'h'},
5182         {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
5183         {"object", required_argument, 0, OPTION_OBJECT},
5184         {"output", required_argument, 0, OPTION_OUTPUT},
5185         {"size", required_argument, 0, OPTION_SIZE},
5186         {"force-share", no_argument, 0, 'U'},
5187         {0, 0, 0, 0}
5188     };
5189     OutputFormat output_format = OFORMAT_HUMAN;
5190     BlockBackend *in_blk = NULL;
5191     BlockDriver *drv;
5192     const char *filename = NULL;
5193     const char *fmt = NULL;
5194     const char *out_fmt = "raw";
5195     char *options = NULL;
5196     char *snapshot_name = NULL;
5197     bool force_share = false;
5198     QemuOpts *opts = NULL;
5199     QemuOpts *object_opts = NULL;
5200     QemuOpts *sn_opts = NULL;
5201     QemuOptsList *create_opts = NULL;
5202     bool image_opts = false;
5203     uint64_t img_size = UINT64_MAX;
5204     BlockMeasureInfo *info = NULL;
5205     Error *local_err = NULL;
5206     int ret = 1;
5207     int c;
5208 
5209     while ((c = getopt_long(argc, argv, "hf:O:o:l:U",
5210                             long_options, NULL)) != -1) {
5211         switch (c) {
5212         case '?':
5213         case 'h':
5214             help();
5215             break;
5216         case 'f':
5217             fmt = optarg;
5218             break;
5219         case 'O':
5220             out_fmt = optarg;
5221             break;
5222         case 'o':
5223             if (accumulate_options(&options, optarg) < 0) {
5224                 goto out;
5225             }
5226             break;
5227         case 'l':
5228             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
5229                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
5230                                                   optarg, false);
5231                 if (!sn_opts) {
5232                     error_report("Failed in parsing snapshot param '%s'",
5233                                  optarg);
5234                     goto out;
5235                 }
5236             } else {
5237                 snapshot_name = optarg;
5238             }
5239             break;
5240         case 'U':
5241             force_share = true;
5242             break;
5243         case OPTION_OBJECT:
5244             user_creatable_process_cmdline(optarg);
5245             break;
5246         case OPTION_IMAGE_OPTS:
5247             image_opts = true;
5248             break;
5249         case OPTION_OUTPUT:
5250             if (!strcmp(optarg, "json")) {
5251                 output_format = OFORMAT_JSON;
5252             } else if (!strcmp(optarg, "human")) {
5253                 output_format = OFORMAT_HUMAN;
5254             } else {
5255                 error_report("--output must be used with human or json "
5256                              "as argument.");
5257                 goto out;
5258             }
5259             break;
5260         case OPTION_SIZE:
5261         {
5262             int64_t sval;
5263 
5264             sval = cvtnum("image size", optarg);
5265             if (sval < 0) {
5266                 goto out;
5267             }
5268             img_size = (uint64_t)sval;
5269         }
5270         break;
5271         }
5272     }
5273 
5274     if (argc - optind > 1) {
5275         error_report("At most one filename argument is allowed.");
5276         goto out;
5277     } else if (argc - optind == 1) {
5278         filename = argv[optind];
5279     }
5280 
5281     if (!filename && (image_opts || fmt || snapshot_name || sn_opts)) {
5282         error_report("--image-opts, -f, and -l require a filename argument.");
5283         goto out;
5284     }
5285     if (filename && img_size != UINT64_MAX) {
5286         error_report("--size N cannot be used together with a filename.");
5287         goto out;
5288     }
5289     if (!filename && img_size == UINT64_MAX) {
5290         error_report("Either --size N or one filename must be specified.");
5291         goto out;
5292     }
5293 
5294     if (filename) {
5295         in_blk = img_open(image_opts, filename, fmt, 0,
5296                           false, false, force_share);
5297         if (!in_blk) {
5298             goto out;
5299         }
5300 
5301         if (sn_opts) {
5302             bdrv_snapshot_load_tmp(blk_bs(in_blk),
5303                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
5304                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
5305                     &local_err);
5306         } else if (snapshot_name != NULL) {
5307             bdrv_snapshot_load_tmp_by_id_or_name(blk_bs(in_blk),
5308                     snapshot_name, &local_err);
5309         }
5310         if (local_err) {
5311             error_reportf_err(local_err, "Failed to load snapshot: ");
5312             goto out;
5313         }
5314     }
5315 
5316     drv = bdrv_find_format(out_fmt);
5317     if (!drv) {
5318         error_report("Unknown file format '%s'", out_fmt);
5319         goto out;
5320     }
5321     if (!drv->create_opts) {
5322         error_report("Format driver '%s' does not support image creation",
5323                      drv->format_name);
5324         goto out;
5325     }
5326 
5327     create_opts = qemu_opts_append(create_opts, drv->create_opts);
5328     create_opts = qemu_opts_append(create_opts, bdrv_file.create_opts);
5329     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5330     if (options) {
5331         if (!qemu_opts_do_parse(opts, options, NULL, &local_err)) {
5332             error_report_err(local_err);
5333             error_report("Invalid options for file format '%s'", out_fmt);
5334             goto out;
5335         }
5336     }
5337     if (img_size != UINT64_MAX) {
5338         qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
5339     }
5340 
5341     info = bdrv_measure(drv, opts, in_blk ? blk_bs(in_blk) : NULL, &local_err);
5342     if (local_err) {
5343         error_report_err(local_err);
5344         goto out;
5345     }
5346 
5347     if (output_format == OFORMAT_HUMAN) {
5348         printf("required size: %" PRIu64 "\n", info->required);
5349         printf("fully allocated size: %" PRIu64 "\n", info->fully_allocated);
5350         if (info->has_bitmaps) {
5351             printf("bitmaps size: %" PRIu64 "\n", info->bitmaps);
5352         }
5353     } else {
5354         dump_json_block_measure_info(info);
5355     }
5356 
5357     ret = 0;
5358 
5359 out:
5360     qapi_free_BlockMeasureInfo(info);
5361     qemu_opts_del(object_opts);
5362     qemu_opts_del(opts);
5363     qemu_opts_del(sn_opts);
5364     qemu_opts_free(create_opts);
5365     g_free(options);
5366     blk_unref(in_blk);
5367     return ret;
5368 }
5369 
5370 static const img_cmd_t img_cmds[] = {
5371 #define DEF(option, callback, arg_string)        \
5372     { option, callback },
5373 #include "qemu-img-cmds.h"
5374 #undef DEF
5375     { NULL, NULL, },
5376 };
5377 
5378 int main(int argc, char **argv)
5379 {
5380     const img_cmd_t *cmd;
5381     const char *cmdname;
5382     int c;
5383     static const struct option long_options[] = {
5384         {"help", no_argument, 0, 'h'},
5385         {"version", no_argument, 0, 'V'},
5386         {"trace", required_argument, NULL, 'T'},
5387         {0, 0, 0, 0}
5388     };
5389 
5390 #ifdef CONFIG_POSIX
5391     signal(SIGPIPE, SIG_IGN);
5392 #endif
5393 
5394     socket_init();
5395     error_init(argv[0]);
5396     module_call_init(MODULE_INIT_TRACE);
5397     qemu_init_exec_dir(argv[0]);
5398 
5399     qemu_init_main_loop(&error_fatal);
5400 
5401     qcrypto_init(&error_fatal);
5402 
5403     module_call_init(MODULE_INIT_QOM);
5404     bdrv_init();
5405     if (argc < 2) {
5406         error_exit("Not enough arguments");
5407     }
5408 
5409     qemu_add_opts(&qemu_source_opts);
5410     qemu_add_opts(&qemu_trace_opts);
5411 
5412     while ((c = getopt_long(argc, argv, "+:hVT:", long_options, NULL)) != -1) {
5413         switch (c) {
5414         case ':':
5415             missing_argument(argv[optind - 1]);
5416             return 0;
5417         case '?':
5418             unrecognized_option(argv[optind - 1]);
5419             return 0;
5420         case 'h':
5421             help();
5422             return 0;
5423         case 'V':
5424             printf(QEMU_IMG_VERSION);
5425             return 0;
5426         case 'T':
5427             trace_opt_parse(optarg);
5428             break;
5429         }
5430     }
5431 
5432     cmdname = argv[optind];
5433 
5434     /* reset getopt_long scanning */
5435     argc -= optind;
5436     if (argc < 1) {
5437         return 0;
5438     }
5439     argv += optind;
5440     qemu_reset_optind();
5441 
5442     if (!trace_init_backends()) {
5443         exit(1);
5444     }
5445     trace_init_file();
5446     qemu_set_log(LOG_TRACE, &error_fatal);
5447 
5448     /* find the command */
5449     for (cmd = img_cmds; cmd->name != NULL; cmd++) {
5450         if (!strcmp(cmdname, cmd->name)) {
5451             return cmd->handler(argc, argv);
5452         }
5453     }
5454 
5455     /* not found */
5456     error_exit("Command not found: %s", cmdname);
5457 }
5458