xref: /openbmc/qemu/qemu-img.c (revision f0830823)
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 
454     if (blk) {
455         blk_set_force_allow_inactivate(blk);
456     }
457 
458     return blk;
459 }
460 
461 
462 static int add_old_style_options(const char *fmt, QemuOpts *opts,
463                                  const char *base_filename,
464                                  const char *base_fmt)
465 {
466     if (base_filename) {
467         if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
468                           NULL)) {
469             error_report("Backing file not supported for file format '%s'",
470                          fmt);
471             return -1;
472         }
473     }
474     if (base_fmt) {
475         if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
476             error_report("Backing file format not supported for file "
477                          "format '%s'", fmt);
478             return -1;
479         }
480     }
481     return 0;
482 }
483 
484 static int64_t cvtnum_full(const char *name, const char *value, int64_t min,
485                            int64_t max)
486 {
487     int err;
488     uint64_t res;
489 
490     err = qemu_strtosz(value, NULL, &res);
491     if (err < 0 && err != -ERANGE) {
492         error_report("Invalid %s specified. You may use "
493                      "k, M, G, T, P or E suffixes for", name);
494         error_report("kilobytes, megabytes, gigabytes, terabytes, "
495                      "petabytes and exabytes.");
496         return err;
497     }
498     if (err == -ERANGE || res > max || res < min) {
499         error_report("Invalid %s specified. Must be between %" PRId64
500                      " and %" PRId64 ".", name, min, max);
501         return -ERANGE;
502     }
503     return res;
504 }
505 
506 static int64_t cvtnum(const char *name, const char *value)
507 {
508     return cvtnum_full(name, value, 0, INT64_MAX);
509 }
510 
511 static int img_create(int argc, char **argv)
512 {
513     int c;
514     uint64_t img_size = -1;
515     const char *fmt = "raw";
516     const char *base_fmt = NULL;
517     const char *filename;
518     const char *base_filename = NULL;
519     char *options = NULL;
520     Error *local_err = NULL;
521     bool quiet = false;
522     int flags = 0;
523 
524     for(;;) {
525         static const struct option long_options[] = {
526             {"help", no_argument, 0, 'h'},
527             {"object", required_argument, 0, OPTION_OBJECT},
528             {0, 0, 0, 0}
529         };
530         c = getopt_long(argc, argv, ":F:b:f:ho:qu",
531                         long_options, NULL);
532         if (c == -1) {
533             break;
534         }
535         switch(c) {
536         case ':':
537             missing_argument(argv[optind - 1]);
538             break;
539         case '?':
540             unrecognized_option(argv[optind - 1]);
541             break;
542         case 'h':
543             help();
544             break;
545         case 'F':
546             base_fmt = optarg;
547             break;
548         case 'b':
549             base_filename = optarg;
550             break;
551         case 'f':
552             fmt = optarg;
553             break;
554         case 'o':
555             if (accumulate_options(&options, optarg) < 0) {
556                 goto fail;
557             }
558             break;
559         case 'q':
560             quiet = true;
561             break;
562         case 'u':
563             flags |= BDRV_O_NO_BACKING;
564             break;
565         case OPTION_OBJECT:
566             user_creatable_process_cmdline(optarg);
567             break;
568         }
569     }
570 
571     /* Get the filename */
572     filename = (optind < argc) ? argv[optind] : NULL;
573     if (options && has_help_option(options)) {
574         g_free(options);
575         return print_block_option_help(filename, fmt);
576     }
577 
578     if (optind >= argc) {
579         error_exit("Expecting image file name");
580     }
581     optind++;
582 
583     /* Get image size, if specified */
584     if (optind < argc) {
585         int64_t sval;
586 
587         sval = cvtnum("image size", argv[optind++]);
588         if (sval < 0) {
589             goto fail;
590         }
591         img_size = (uint64_t)sval;
592     }
593     if (optind != argc) {
594         error_exit("Unexpected argument: %s", argv[optind]);
595     }
596 
597     bdrv_img_create(filename, fmt, base_filename, base_fmt,
598                     options, img_size, flags, quiet, &local_err);
599     if (local_err) {
600         error_reportf_err(local_err, "%s: ", filename);
601         goto fail;
602     }
603 
604     g_free(options);
605     return 0;
606 
607 fail:
608     g_free(options);
609     return 1;
610 }
611 
612 static void dump_json_image_check(ImageCheck *check, bool quiet)
613 {
614     GString *str;
615     QObject *obj;
616     Visitor *v = qobject_output_visitor_new(&obj);
617 
618     visit_type_ImageCheck(v, NULL, &check, &error_abort);
619     visit_complete(v, &obj);
620     str = qobject_to_json_pretty(obj, true);
621     assert(str != NULL);
622     qprintf(quiet, "%s\n", str->str);
623     qobject_unref(obj);
624     visit_free(v);
625     g_string_free(str, true);
626 }
627 
628 static void dump_human_image_check(ImageCheck *check, bool quiet)
629 {
630     if (!(check->corruptions || check->leaks || check->check_errors)) {
631         qprintf(quiet, "No errors were found on the image.\n");
632     } else {
633         if (check->corruptions) {
634             qprintf(quiet, "\n%" PRId64 " errors were found on the image.\n"
635                     "Data may be corrupted, or further writes to the image "
636                     "may corrupt it.\n",
637                     check->corruptions);
638         }
639 
640         if (check->leaks) {
641             qprintf(quiet,
642                     "\n%" PRId64 " leaked clusters were found on the image.\n"
643                     "This means waste of disk space, but no harm to data.\n",
644                     check->leaks);
645         }
646 
647         if (check->check_errors) {
648             qprintf(quiet,
649                     "\n%" PRId64
650                     " internal errors have occurred during the check.\n",
651                     check->check_errors);
652         }
653     }
654 
655     if (check->total_clusters != 0 && check->allocated_clusters != 0) {
656         qprintf(quiet, "%" PRId64 "/%" PRId64 " = %0.2f%% allocated, "
657                 "%0.2f%% fragmented, %0.2f%% compressed clusters\n",
658                 check->allocated_clusters, check->total_clusters,
659                 check->allocated_clusters * 100.0 / check->total_clusters,
660                 check->fragmented_clusters * 100.0 / check->allocated_clusters,
661                 check->compressed_clusters * 100.0 /
662                 check->allocated_clusters);
663     }
664 
665     if (check->image_end_offset) {
666         qprintf(quiet,
667                 "Image end offset: %" PRId64 "\n", check->image_end_offset);
668     }
669 }
670 
671 static int collect_image_check(BlockDriverState *bs,
672                    ImageCheck *check,
673                    const char *filename,
674                    const char *fmt,
675                    int fix)
676 {
677     int ret;
678     BdrvCheckResult result;
679 
680     ret = bdrv_check(bs, &result, fix);
681     if (ret < 0) {
682         return ret;
683     }
684 
685     check->filename                 = g_strdup(filename);
686     check->format                   = g_strdup(bdrv_get_format_name(bs));
687     check->check_errors             = result.check_errors;
688     check->corruptions              = result.corruptions;
689     check->has_corruptions          = result.corruptions != 0;
690     check->leaks                    = result.leaks;
691     check->has_leaks                = result.leaks != 0;
692     check->corruptions_fixed        = result.corruptions_fixed;
693     check->has_corruptions_fixed    = result.corruptions_fixed != 0;
694     check->leaks_fixed              = result.leaks_fixed;
695     check->has_leaks_fixed          = result.leaks_fixed != 0;
696     check->image_end_offset         = result.image_end_offset;
697     check->has_image_end_offset     = result.image_end_offset != 0;
698     check->total_clusters           = result.bfi.total_clusters;
699     check->has_total_clusters       = result.bfi.total_clusters != 0;
700     check->allocated_clusters       = result.bfi.allocated_clusters;
701     check->has_allocated_clusters   = result.bfi.allocated_clusters != 0;
702     check->fragmented_clusters      = result.bfi.fragmented_clusters;
703     check->has_fragmented_clusters  = result.bfi.fragmented_clusters != 0;
704     check->compressed_clusters      = result.bfi.compressed_clusters;
705     check->has_compressed_clusters  = result.bfi.compressed_clusters != 0;
706 
707     return 0;
708 }
709 
710 /*
711  * Checks an image for consistency. Exit codes:
712  *
713  *  0 - Check completed, image is good
714  *  1 - Check not completed because of internal errors
715  *  2 - Check completed, image is corrupted
716  *  3 - Check completed, image has leaked clusters, but is good otherwise
717  * 63 - Checks are not supported by the image format
718  */
719 static int img_check(int argc, char **argv)
720 {
721     int c, ret;
722     OutputFormat output_format = OFORMAT_HUMAN;
723     const char *filename, *fmt, *output, *cache;
724     BlockBackend *blk;
725     BlockDriverState *bs;
726     int fix = 0;
727     int flags = BDRV_O_CHECK;
728     bool writethrough;
729     ImageCheck *check;
730     bool quiet = false;
731     bool image_opts = false;
732     bool force_share = false;
733 
734     fmt = NULL;
735     output = NULL;
736     cache = BDRV_DEFAULT_CACHE;
737 
738     for(;;) {
739         int option_index = 0;
740         static const struct option long_options[] = {
741             {"help", no_argument, 0, 'h'},
742             {"format", required_argument, 0, 'f'},
743             {"repair", required_argument, 0, 'r'},
744             {"output", required_argument, 0, OPTION_OUTPUT},
745             {"object", required_argument, 0, OPTION_OBJECT},
746             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
747             {"force-share", no_argument, 0, 'U'},
748             {0, 0, 0, 0}
749         };
750         c = getopt_long(argc, argv, ":hf:r:T:qU",
751                         long_options, &option_index);
752         if (c == -1) {
753             break;
754         }
755         switch(c) {
756         case ':':
757             missing_argument(argv[optind - 1]);
758             break;
759         case '?':
760             unrecognized_option(argv[optind - 1]);
761             break;
762         case 'h':
763             help();
764             break;
765         case 'f':
766             fmt = optarg;
767             break;
768         case 'r':
769             flags |= BDRV_O_RDWR;
770 
771             if (!strcmp(optarg, "leaks")) {
772                 fix = BDRV_FIX_LEAKS;
773             } else if (!strcmp(optarg, "all")) {
774                 fix = BDRV_FIX_LEAKS | BDRV_FIX_ERRORS;
775             } else {
776                 error_exit("Unknown option value for -r "
777                            "(expecting 'leaks' or 'all'): %s", optarg);
778             }
779             break;
780         case OPTION_OUTPUT:
781             output = optarg;
782             break;
783         case 'T':
784             cache = optarg;
785             break;
786         case 'q':
787             quiet = true;
788             break;
789         case 'U':
790             force_share = true;
791             break;
792         case OPTION_OBJECT:
793             user_creatable_process_cmdline(optarg);
794             break;
795         case OPTION_IMAGE_OPTS:
796             image_opts = true;
797             break;
798         }
799     }
800     if (optind != argc - 1) {
801         error_exit("Expecting one image file name");
802     }
803     filename = argv[optind++];
804 
805     if (output && !strcmp(output, "json")) {
806         output_format = OFORMAT_JSON;
807     } else if (output && !strcmp(output, "human")) {
808         output_format = OFORMAT_HUMAN;
809     } else if (output) {
810         error_report("--output must be used with human or json as argument.");
811         return 1;
812     }
813 
814     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
815     if (ret < 0) {
816         error_report("Invalid source cache option: %s", cache);
817         return 1;
818     }
819 
820     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
821                    force_share);
822     if (!blk) {
823         return 1;
824     }
825     bs = blk_bs(blk);
826 
827     check = g_new0(ImageCheck, 1);
828     ret = collect_image_check(bs, check, filename, fmt, fix);
829 
830     if (ret == -ENOTSUP) {
831         error_report("This image format does not support checks");
832         ret = 63;
833         goto fail;
834     }
835 
836     if (check->corruptions_fixed || check->leaks_fixed) {
837         int corruptions_fixed, leaks_fixed;
838         bool has_leaks_fixed, has_corruptions_fixed;
839 
840         leaks_fixed         = check->leaks_fixed;
841         has_leaks_fixed     = check->has_leaks_fixed;
842         corruptions_fixed   = check->corruptions_fixed;
843         has_corruptions_fixed = check->has_corruptions_fixed;
844 
845         if (output_format == OFORMAT_HUMAN) {
846             qprintf(quiet,
847                     "The following inconsistencies were found and repaired:\n\n"
848                     "    %" PRId64 " leaked clusters\n"
849                     "    %" PRId64 " corruptions\n\n"
850                     "Double checking the fixed image now...\n",
851                     check->leaks_fixed,
852                     check->corruptions_fixed);
853         }
854 
855         qapi_free_ImageCheck(check);
856         check = g_new0(ImageCheck, 1);
857         ret = collect_image_check(bs, check, filename, fmt, 0);
858 
859         check->leaks_fixed          = leaks_fixed;
860         check->has_leaks_fixed      = has_leaks_fixed;
861         check->corruptions_fixed    = corruptions_fixed;
862         check->has_corruptions_fixed = has_corruptions_fixed;
863     }
864 
865     if (!ret) {
866         switch (output_format) {
867         case OFORMAT_HUMAN:
868             dump_human_image_check(check, quiet);
869             break;
870         case OFORMAT_JSON:
871             dump_json_image_check(check, quiet);
872             break;
873         }
874     }
875 
876     if (ret || check->check_errors) {
877         if (ret) {
878             error_report("Check failed: %s", strerror(-ret));
879         } else {
880             error_report("Check failed");
881         }
882         ret = 1;
883         goto fail;
884     }
885 
886     if (check->corruptions) {
887         ret = 2;
888     } else if (check->leaks) {
889         ret = 3;
890     } else {
891         ret = 0;
892     }
893 
894 fail:
895     qapi_free_ImageCheck(check);
896     blk_unref(blk);
897     return ret;
898 }
899 
900 typedef struct CommonBlockJobCBInfo {
901     BlockDriverState *bs;
902     Error **errp;
903 } CommonBlockJobCBInfo;
904 
905 static void common_block_job_cb(void *opaque, int ret)
906 {
907     CommonBlockJobCBInfo *cbi = opaque;
908 
909     if (ret < 0) {
910         error_setg_errno(cbi->errp, -ret, "Block job failed");
911     }
912 }
913 
914 static void run_block_job(BlockJob *job, Error **errp)
915 {
916     uint64_t progress_current, progress_total;
917     AioContext *aio_context = block_job_get_aio_context(job);
918     int ret = 0;
919 
920     job_lock();
921     job_ref_locked(&job->job);
922     do {
923         float progress = 0.0f;
924         job_unlock();
925         aio_poll(aio_context, true);
926 
927         progress_get_snapshot(&job->job.progress, &progress_current,
928                               &progress_total);
929         if (progress_total) {
930             progress = (float)progress_current / progress_total * 100.f;
931         }
932         qemu_progress_print(progress, 0);
933         job_lock();
934     } while (!job_is_ready_locked(&job->job) &&
935              !job_is_completed_locked(&job->job));
936 
937     if (!job_is_completed_locked(&job->job)) {
938         ret = job_complete_sync_locked(&job->job, errp);
939     } else {
940         ret = job->job.ret;
941     }
942     job_unref_locked(&job->job);
943     job_unlock();
944 
945     /* publish completion progress only when success */
946     if (!ret) {
947         qemu_progress_print(100.f, 0);
948     }
949 }
950 
951 static int img_commit(int argc, char **argv)
952 {
953     int c, ret, flags;
954     const char *filename, *fmt, *cache, *base;
955     BlockBackend *blk;
956     BlockDriverState *bs, *base_bs;
957     BlockJob *job;
958     bool progress = false, quiet = false, drop = false;
959     bool writethrough;
960     Error *local_err = NULL;
961     CommonBlockJobCBInfo cbi;
962     bool image_opts = false;
963     AioContext *aio_context;
964     int64_t rate_limit = 0;
965 
966     fmt = NULL;
967     cache = BDRV_DEFAULT_CACHE;
968     base = NULL;
969     for(;;) {
970         static const struct option long_options[] = {
971             {"help", no_argument, 0, 'h'},
972             {"object", required_argument, 0, OPTION_OBJECT},
973             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
974             {0, 0, 0, 0}
975         };
976         c = getopt_long(argc, argv, ":f:ht:b:dpqr:",
977                         long_options, NULL);
978         if (c == -1) {
979             break;
980         }
981         switch(c) {
982         case ':':
983             missing_argument(argv[optind - 1]);
984             break;
985         case '?':
986             unrecognized_option(argv[optind - 1]);
987             break;
988         case 'h':
989             help();
990             break;
991         case 'f':
992             fmt = optarg;
993             break;
994         case 't':
995             cache = optarg;
996             break;
997         case 'b':
998             base = optarg;
999             /* -b implies -d */
1000             drop = true;
1001             break;
1002         case 'd':
1003             drop = true;
1004             break;
1005         case 'p':
1006             progress = true;
1007             break;
1008         case 'q':
1009             quiet = true;
1010             break;
1011         case 'r':
1012             rate_limit = cvtnum("rate limit", optarg);
1013             if (rate_limit < 0) {
1014                 return 1;
1015             }
1016             break;
1017         case OPTION_OBJECT:
1018             user_creatable_process_cmdline(optarg);
1019             break;
1020         case OPTION_IMAGE_OPTS:
1021             image_opts = true;
1022             break;
1023         }
1024     }
1025 
1026     /* Progress is not shown in Quiet mode */
1027     if (quiet) {
1028         progress = false;
1029     }
1030 
1031     if (optind != argc - 1) {
1032         error_exit("Expecting one image file name");
1033     }
1034     filename = argv[optind++];
1035 
1036     flags = BDRV_O_RDWR | BDRV_O_UNMAP;
1037     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
1038     if (ret < 0) {
1039         error_report("Invalid cache option: %s", cache);
1040         return 1;
1041     }
1042 
1043     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
1044                    false);
1045     if (!blk) {
1046         return 1;
1047     }
1048     bs = blk_bs(blk);
1049 
1050     qemu_progress_init(progress, 1.f);
1051     qemu_progress_print(0.f, 100);
1052 
1053     if (base) {
1054         base_bs = bdrv_find_backing_image(bs, base);
1055         if (!base_bs) {
1056             error_setg(&local_err,
1057                        "Did not find '%s' in the backing chain of '%s'",
1058                        base, filename);
1059             goto done;
1060         }
1061     } else {
1062         /* This is different from QMP, which by default uses the deepest file in
1063          * the backing chain (i.e., the very base); however, the traditional
1064          * behavior of qemu-img commit is using the immediate backing file. */
1065         base_bs = bdrv_backing_chain_next(bs);
1066         if (!base_bs) {
1067             error_setg(&local_err, "Image does not have a backing file");
1068             goto done;
1069         }
1070     }
1071 
1072     cbi = (CommonBlockJobCBInfo){
1073         .errp = &local_err,
1074         .bs   = bs,
1075     };
1076 
1077     aio_context = bdrv_get_aio_context(bs);
1078     aio_context_acquire(aio_context);
1079     commit_active_start("commit", bs, base_bs, JOB_DEFAULT, rate_limit,
1080                         BLOCKDEV_ON_ERROR_REPORT, NULL, common_block_job_cb,
1081                         &cbi, false, &local_err);
1082     aio_context_release(aio_context);
1083     if (local_err) {
1084         goto done;
1085     }
1086 
1087     /* When the block job completes, the BlockBackend reference will point to
1088      * the old backing file. In order to avoid that the top image is already
1089      * deleted, so we can still empty it afterwards, increment the reference
1090      * counter here preemptively. */
1091     if (!drop) {
1092         bdrv_ref(bs);
1093     }
1094 
1095     job = block_job_get("commit");
1096     assert(job);
1097     run_block_job(job, &local_err);
1098     if (local_err) {
1099         goto unref_backing;
1100     }
1101 
1102     if (!drop) {
1103         BlockBackend *old_backing_blk;
1104 
1105         old_backing_blk = blk_new_with_bs(bs, BLK_PERM_WRITE, BLK_PERM_ALL,
1106                                           &local_err);
1107         if (!old_backing_blk) {
1108             goto unref_backing;
1109         }
1110         ret = blk_make_empty(old_backing_blk, &local_err);
1111         blk_unref(old_backing_blk);
1112         if (ret == -ENOTSUP) {
1113             error_free(local_err);
1114             local_err = NULL;
1115         } else if (ret < 0) {
1116             goto unref_backing;
1117         }
1118     }
1119 
1120 unref_backing:
1121     if (!drop) {
1122         bdrv_unref(bs);
1123     }
1124 
1125 done:
1126     qemu_progress_end();
1127 
1128     /*
1129      * Manually inactivate the image first because this way we can know whether
1130      * an error occurred. blk_unref() doesn't tell us about failures.
1131      */
1132     ret = bdrv_inactivate_all();
1133     if (ret < 0 && !local_err) {
1134         error_setg_errno(&local_err, -ret, "Error while closing the image");
1135     }
1136     blk_unref(blk);
1137 
1138     if (local_err) {
1139         error_report_err(local_err);
1140         return 1;
1141     }
1142 
1143     qprintf(quiet, "Image committed.\n");
1144     return 0;
1145 }
1146 
1147 /*
1148  * Returns -1 if 'buf' contains only zeroes, otherwise the byte index
1149  * of the first sector boundary within buf where the sector contains a
1150  * non-zero byte.  This function is robust to a buffer that is not
1151  * sector-aligned.
1152  */
1153 static int64_t find_nonzero(const uint8_t *buf, int64_t n)
1154 {
1155     int64_t i;
1156     int64_t end = QEMU_ALIGN_DOWN(n, BDRV_SECTOR_SIZE);
1157 
1158     for (i = 0; i < end; i += BDRV_SECTOR_SIZE) {
1159         if (!buffer_is_zero(buf + i, BDRV_SECTOR_SIZE)) {
1160             return i;
1161         }
1162     }
1163     if (i < n && !buffer_is_zero(buf + i, n - end)) {
1164         return i;
1165     }
1166     return -1;
1167 }
1168 
1169 /*
1170  * Returns true iff the first sector pointed to by 'buf' contains at least
1171  * a non-NUL byte.
1172  *
1173  * 'pnum' is set to the number of sectors (including and immediately following
1174  * the first one) that are known to be in the same allocated/unallocated state.
1175  * The function will try to align the end offset to alignment boundaries so
1176  * that the request will at least end aligned and consecutive requests will
1177  * also start at an aligned offset.
1178  */
1179 static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum,
1180                                 int64_t sector_num, int alignment)
1181 {
1182     bool is_zero;
1183     int i, tail;
1184 
1185     if (n <= 0) {
1186         *pnum = 0;
1187         return 0;
1188     }
1189     is_zero = buffer_is_zero(buf, BDRV_SECTOR_SIZE);
1190     for(i = 1; i < n; i++) {
1191         buf += BDRV_SECTOR_SIZE;
1192         if (is_zero != buffer_is_zero(buf, BDRV_SECTOR_SIZE)) {
1193             break;
1194         }
1195     }
1196 
1197     if (i == n) {
1198         /*
1199          * The whole buf is the same.
1200          * No reason to split it into chunks, so return now.
1201          */
1202         *pnum = i;
1203         return !is_zero;
1204     }
1205 
1206     tail = (sector_num + i) & (alignment - 1);
1207     if (tail) {
1208         if (is_zero && i <= tail) {
1209             /*
1210              * For sure next sector after i is data, and it will rewrite this
1211              * tail anyway due to RMW. So, let's just write data now.
1212              */
1213             is_zero = false;
1214         }
1215         if (!is_zero) {
1216             /* If possible, align up end offset of allocated areas. */
1217             i += alignment - tail;
1218             i = MIN(i, n);
1219         } else {
1220             /*
1221              * For sure next sector after i is data, and it will rewrite this
1222              * tail anyway due to RMW. Better is avoid RMW and write zeroes up
1223              * to aligned bound.
1224              */
1225             i -= tail;
1226         }
1227     }
1228     *pnum = i;
1229     return !is_zero;
1230 }
1231 
1232 /*
1233  * Like is_allocated_sectors, but if the buffer starts with a used sector,
1234  * up to 'min' consecutive sectors containing zeros are ignored. This avoids
1235  * breaking up write requests for only small sparse areas.
1236  */
1237 static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum,
1238     int min, int64_t sector_num, int alignment)
1239 {
1240     int ret;
1241     int num_checked, num_used;
1242 
1243     if (n < min) {
1244         min = n;
1245     }
1246 
1247     ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
1248     if (!ret) {
1249         return ret;
1250     }
1251 
1252     num_used = *pnum;
1253     buf += BDRV_SECTOR_SIZE * *pnum;
1254     n -= *pnum;
1255     sector_num += *pnum;
1256     num_checked = num_used;
1257 
1258     while (n > 0) {
1259         ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
1260 
1261         buf += BDRV_SECTOR_SIZE * *pnum;
1262         n -= *pnum;
1263         sector_num += *pnum;
1264         num_checked += *pnum;
1265         if (ret) {
1266             num_used = num_checked;
1267         } else if (*pnum >= min) {
1268             break;
1269         }
1270     }
1271 
1272     *pnum = num_used;
1273     return 1;
1274 }
1275 
1276 /*
1277  * Compares two buffers sector by sector. Returns 0 if the first
1278  * sector of each buffer matches, non-zero otherwise.
1279  *
1280  * pnum is set to the sector-aligned size of the buffer prefix that
1281  * has the same matching status as the first sector.
1282  */
1283 static int compare_buffers(const uint8_t *buf1, const uint8_t *buf2,
1284                            int64_t bytes, int64_t *pnum)
1285 {
1286     bool res;
1287     int64_t i = MIN(bytes, BDRV_SECTOR_SIZE);
1288 
1289     assert(bytes > 0);
1290 
1291     res = !!memcmp(buf1, buf2, i);
1292     while (i < bytes) {
1293         int64_t len = MIN(bytes - i, BDRV_SECTOR_SIZE);
1294 
1295         if (!!memcmp(buf1 + i, buf2 + i, len) != res) {
1296             break;
1297         }
1298         i += len;
1299     }
1300 
1301     *pnum = i;
1302     return res;
1303 }
1304 
1305 #define IO_BUF_SIZE (2 * MiB)
1306 
1307 /*
1308  * Check if passed sectors are empty (not allocated or contain only 0 bytes)
1309  *
1310  * Intended for use by 'qemu-img compare': Returns 0 in case sectors are
1311  * filled with 0, 1 if sectors contain non-zero data (this is a comparison
1312  * failure), and 4 on error (the exit status for read errors), after emitting
1313  * an error message.
1314  *
1315  * @param blk:  BlockBackend for the image
1316  * @param offset: Starting offset to check
1317  * @param bytes: Number of bytes to check
1318  * @param filename: Name of disk file we are checking (logging purpose)
1319  * @param buffer: Allocated buffer for storing read data
1320  * @param quiet: Flag for quiet mode
1321  */
1322 static int check_empty_sectors(BlockBackend *blk, int64_t offset,
1323                                int64_t bytes, const char *filename,
1324                                uint8_t *buffer, bool quiet)
1325 {
1326     int ret = 0;
1327     int64_t idx;
1328 
1329     ret = blk_pread(blk, offset, bytes, buffer, 0);
1330     if (ret < 0) {
1331         error_report("Error while reading offset %" PRId64 " of %s: %s",
1332                      offset, filename, strerror(-ret));
1333         return 4;
1334     }
1335     idx = find_nonzero(buffer, bytes);
1336     if (idx >= 0) {
1337         qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1338                 offset + idx);
1339         return 1;
1340     }
1341 
1342     return 0;
1343 }
1344 
1345 /*
1346  * Compares two images. Exit codes:
1347  *
1348  * 0 - Images are identical or the requested help was printed
1349  * 1 - Images differ
1350  * >1 - Error occurred
1351  */
1352 static int img_compare(int argc, char **argv)
1353 {
1354     const char *fmt1 = NULL, *fmt2 = NULL, *cache, *filename1, *filename2;
1355     BlockBackend *blk1, *blk2;
1356     BlockDriverState *bs1, *bs2;
1357     int64_t total_size1, total_size2;
1358     uint8_t *buf1 = NULL, *buf2 = NULL;
1359     int64_t pnum1, pnum2;
1360     int allocated1, allocated2;
1361     int ret = 0; /* return value - 0 Ident, 1 Different, >1 Error */
1362     bool progress = false, quiet = false, strict = false;
1363     int flags;
1364     bool writethrough;
1365     int64_t total_size;
1366     int64_t offset = 0;
1367     int64_t chunk;
1368     int c;
1369     uint64_t progress_base;
1370     bool image_opts = false;
1371     bool force_share = false;
1372 
1373     cache = BDRV_DEFAULT_CACHE;
1374     for (;;) {
1375         static const struct option long_options[] = {
1376             {"help", no_argument, 0, 'h'},
1377             {"object", required_argument, 0, OPTION_OBJECT},
1378             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
1379             {"force-share", no_argument, 0, 'U'},
1380             {0, 0, 0, 0}
1381         };
1382         c = getopt_long(argc, argv, ":hf:F:T:pqsU",
1383                         long_options, NULL);
1384         if (c == -1) {
1385             break;
1386         }
1387         switch (c) {
1388         case ':':
1389             missing_argument(argv[optind - 1]);
1390             break;
1391         case '?':
1392             unrecognized_option(argv[optind - 1]);
1393             break;
1394         case 'h':
1395             help();
1396             break;
1397         case 'f':
1398             fmt1 = optarg;
1399             break;
1400         case 'F':
1401             fmt2 = optarg;
1402             break;
1403         case 'T':
1404             cache = optarg;
1405             break;
1406         case 'p':
1407             progress = true;
1408             break;
1409         case 'q':
1410             quiet = true;
1411             break;
1412         case 's':
1413             strict = true;
1414             break;
1415         case 'U':
1416             force_share = true;
1417             break;
1418         case OPTION_OBJECT:
1419             {
1420                 Error *local_err = NULL;
1421 
1422                 if (!user_creatable_add_from_str(optarg, &local_err)) {
1423                     if (local_err) {
1424                         error_report_err(local_err);
1425                         exit(2);
1426                     } else {
1427                         /* Help was printed */
1428                         exit(EXIT_SUCCESS);
1429                     }
1430                 }
1431                 break;
1432             }
1433         case OPTION_IMAGE_OPTS:
1434             image_opts = true;
1435             break;
1436         }
1437     }
1438 
1439     /* Progress is not shown in Quiet mode */
1440     if (quiet) {
1441         progress = false;
1442     }
1443 
1444 
1445     if (optind != argc - 2) {
1446         error_exit("Expecting two image file names");
1447     }
1448     filename1 = argv[optind++];
1449     filename2 = argv[optind++];
1450 
1451     /* Initialize before goto out */
1452     qemu_progress_init(progress, 2.0);
1453 
1454     flags = 0;
1455     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
1456     if (ret < 0) {
1457         error_report("Invalid source cache option: %s", cache);
1458         ret = 2;
1459         goto out3;
1460     }
1461 
1462     blk1 = img_open(image_opts, filename1, fmt1, flags, writethrough, quiet,
1463                     force_share);
1464     if (!blk1) {
1465         ret = 2;
1466         goto out3;
1467     }
1468 
1469     blk2 = img_open(image_opts, filename2, fmt2, flags, writethrough, quiet,
1470                     force_share);
1471     if (!blk2) {
1472         ret = 2;
1473         goto out2;
1474     }
1475     bs1 = blk_bs(blk1);
1476     bs2 = blk_bs(blk2);
1477 
1478     buf1 = blk_blockalign(blk1, IO_BUF_SIZE);
1479     buf2 = blk_blockalign(blk2, IO_BUF_SIZE);
1480     total_size1 = blk_getlength(blk1);
1481     if (total_size1 < 0) {
1482         error_report("Can't get size of %s: %s",
1483                      filename1, strerror(-total_size1));
1484         ret = 4;
1485         goto out;
1486     }
1487     total_size2 = blk_getlength(blk2);
1488     if (total_size2 < 0) {
1489         error_report("Can't get size of %s: %s",
1490                      filename2, strerror(-total_size2));
1491         ret = 4;
1492         goto out;
1493     }
1494     total_size = MIN(total_size1, total_size2);
1495     progress_base = MAX(total_size1, total_size2);
1496 
1497     qemu_progress_print(0, 100);
1498 
1499     if (strict && total_size1 != total_size2) {
1500         ret = 1;
1501         qprintf(quiet, "Strict mode: Image size mismatch!\n");
1502         goto out;
1503     }
1504 
1505     while (offset < total_size) {
1506         int status1, status2;
1507 
1508         status1 = bdrv_block_status_above(bs1, NULL, offset,
1509                                           total_size1 - offset, &pnum1, NULL,
1510                                           NULL);
1511         if (status1 < 0) {
1512             ret = 3;
1513             error_report("Sector allocation test failed for %s", filename1);
1514             goto out;
1515         }
1516         allocated1 = status1 & BDRV_BLOCK_ALLOCATED;
1517 
1518         status2 = bdrv_block_status_above(bs2, NULL, offset,
1519                                           total_size2 - offset, &pnum2, NULL,
1520                                           NULL);
1521         if (status2 < 0) {
1522             ret = 3;
1523             error_report("Sector allocation test failed for %s", filename2);
1524             goto out;
1525         }
1526         allocated2 = status2 & BDRV_BLOCK_ALLOCATED;
1527 
1528         assert(pnum1 && pnum2);
1529         chunk = MIN(pnum1, pnum2);
1530 
1531         if (strict) {
1532             if (status1 != status2) {
1533                 ret = 1;
1534                 qprintf(quiet, "Strict mode: Offset %" PRId64
1535                         " block status mismatch!\n", offset);
1536                 goto out;
1537             }
1538         }
1539         if ((status1 & BDRV_BLOCK_ZERO) && (status2 & BDRV_BLOCK_ZERO)) {
1540             /* nothing to do */
1541         } else if (allocated1 == allocated2) {
1542             if (allocated1) {
1543                 int64_t pnum;
1544 
1545                 chunk = MIN(chunk, IO_BUF_SIZE);
1546                 ret = blk_pread(blk1, offset, chunk, buf1, 0);
1547                 if (ret < 0) {
1548                     error_report("Error while reading offset %" PRId64
1549                                  " of %s: %s",
1550                                  offset, filename1, strerror(-ret));
1551                     ret = 4;
1552                     goto out;
1553                 }
1554                 ret = blk_pread(blk2, offset, chunk, buf2, 0);
1555                 if (ret < 0) {
1556                     error_report("Error while reading offset %" PRId64
1557                                  " of %s: %s",
1558                                  offset, filename2, strerror(-ret));
1559                     ret = 4;
1560                     goto out;
1561                 }
1562                 ret = compare_buffers(buf1, buf2, chunk, &pnum);
1563                 if (ret || pnum != chunk) {
1564                     qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1565                             offset + (ret ? 0 : pnum));
1566                     ret = 1;
1567                     goto out;
1568                 }
1569             }
1570         } else {
1571             chunk = MIN(chunk, IO_BUF_SIZE);
1572             if (allocated1) {
1573                 ret = check_empty_sectors(blk1, offset, chunk,
1574                                           filename1, buf1, quiet);
1575             } else {
1576                 ret = check_empty_sectors(blk2, offset, chunk,
1577                                           filename2, buf1, quiet);
1578             }
1579             if (ret) {
1580                 goto out;
1581             }
1582         }
1583         offset += chunk;
1584         qemu_progress_print(((float) chunk / progress_base) * 100, 100);
1585     }
1586 
1587     if (total_size1 != total_size2) {
1588         BlockBackend *blk_over;
1589         const char *filename_over;
1590 
1591         qprintf(quiet, "Warning: Image size mismatch!\n");
1592         if (total_size1 > total_size2) {
1593             blk_over = blk1;
1594             filename_over = filename1;
1595         } else {
1596             blk_over = blk2;
1597             filename_over = filename2;
1598         }
1599 
1600         while (offset < progress_base) {
1601             ret = bdrv_block_status_above(blk_bs(blk_over), NULL, offset,
1602                                           progress_base - offset, &chunk,
1603                                           NULL, NULL);
1604             if (ret < 0) {
1605                 ret = 3;
1606                 error_report("Sector allocation test failed for %s",
1607                              filename_over);
1608                 goto out;
1609 
1610             }
1611             if (ret & BDRV_BLOCK_ALLOCATED && !(ret & BDRV_BLOCK_ZERO)) {
1612                 chunk = MIN(chunk, IO_BUF_SIZE);
1613                 ret = check_empty_sectors(blk_over, offset, chunk,
1614                                           filename_over, buf1, quiet);
1615                 if (ret) {
1616                     goto out;
1617                 }
1618             }
1619             offset += chunk;
1620             qemu_progress_print(((float) chunk / progress_base) * 100, 100);
1621         }
1622     }
1623 
1624     qprintf(quiet, "Images are identical.\n");
1625     ret = 0;
1626 
1627 out:
1628     qemu_vfree(buf1);
1629     qemu_vfree(buf2);
1630     blk_unref(blk2);
1631 out2:
1632     blk_unref(blk1);
1633 out3:
1634     qemu_progress_end();
1635     return ret;
1636 }
1637 
1638 /* Convenience wrapper around qmp_block_dirty_bitmap_merge */
1639 static void do_dirty_bitmap_merge(const char *dst_node, const char *dst_name,
1640                                   const char *src_node, const char *src_name,
1641                                   Error **errp)
1642 {
1643     BlockDirtyBitmapOrStr *merge_src;
1644     BlockDirtyBitmapOrStrList *list = NULL;
1645 
1646     merge_src = g_new0(BlockDirtyBitmapOrStr, 1);
1647     merge_src->type = QTYPE_QDICT;
1648     merge_src->u.external.node = g_strdup(src_node);
1649     merge_src->u.external.name = g_strdup(src_name);
1650     QAPI_LIST_PREPEND(list, merge_src);
1651     qmp_block_dirty_bitmap_merge(dst_node, dst_name, list, errp);
1652     qapi_free_BlockDirtyBitmapOrStrList(list);
1653 }
1654 
1655 enum ImgConvertBlockStatus {
1656     BLK_DATA,
1657     BLK_ZERO,
1658     BLK_BACKING_FILE,
1659 };
1660 
1661 #define MAX_COROUTINES 16
1662 #define CONVERT_THROTTLE_GROUP "img_convert"
1663 
1664 typedef struct ImgConvertState {
1665     BlockBackend **src;
1666     int64_t *src_sectors;
1667     int *src_alignment;
1668     int src_num;
1669     int64_t total_sectors;
1670     int64_t allocated_sectors;
1671     int64_t allocated_done;
1672     int64_t sector_num;
1673     int64_t wr_offs;
1674     enum ImgConvertBlockStatus status;
1675     int64_t sector_next_status;
1676     BlockBackend *target;
1677     bool has_zero_init;
1678     bool compressed;
1679     bool target_is_new;
1680     bool target_has_backing;
1681     int64_t target_backing_sectors; /* negative if unknown */
1682     bool wr_in_order;
1683     bool copy_range;
1684     bool salvage;
1685     bool quiet;
1686     int min_sparse;
1687     int alignment;
1688     size_t cluster_sectors;
1689     size_t buf_sectors;
1690     long num_coroutines;
1691     int running_coroutines;
1692     Coroutine *co[MAX_COROUTINES];
1693     int64_t wait_sector_num[MAX_COROUTINES];
1694     CoMutex lock;
1695     int ret;
1696 } ImgConvertState;
1697 
1698 static void convert_select_part(ImgConvertState *s, int64_t sector_num,
1699                                 int *src_cur, int64_t *src_cur_offset)
1700 {
1701     *src_cur = 0;
1702     *src_cur_offset = 0;
1703     while (sector_num - *src_cur_offset >= s->src_sectors[*src_cur]) {
1704         *src_cur_offset += s->src_sectors[*src_cur];
1705         (*src_cur)++;
1706         assert(*src_cur < s->src_num);
1707     }
1708 }
1709 
1710 static int convert_iteration_sectors(ImgConvertState *s, int64_t sector_num)
1711 {
1712     int64_t src_cur_offset;
1713     int ret, n, src_cur;
1714     bool post_backing_zero = false;
1715 
1716     convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1717 
1718     assert(s->total_sectors > sector_num);
1719     n = MIN(s->total_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
1720 
1721     if (s->target_backing_sectors >= 0) {
1722         if (sector_num >= s->target_backing_sectors) {
1723             post_backing_zero = true;
1724         } else if (sector_num + n > s->target_backing_sectors) {
1725             /* Split requests around target_backing_sectors (because
1726              * starting from there, zeros are handled differently) */
1727             n = s->target_backing_sectors - sector_num;
1728         }
1729     }
1730 
1731     if (s->sector_next_status <= sector_num) {
1732         uint64_t offset = (sector_num - src_cur_offset) * BDRV_SECTOR_SIZE;
1733         int64_t count;
1734         int tail;
1735         BlockDriverState *src_bs = blk_bs(s->src[src_cur]);
1736         BlockDriverState *base;
1737 
1738         if (s->target_has_backing) {
1739             base = bdrv_cow_bs(bdrv_skip_filters(src_bs));
1740         } else {
1741             base = NULL;
1742         }
1743 
1744         do {
1745             count = n * BDRV_SECTOR_SIZE;
1746 
1747             ret = bdrv_block_status_above(src_bs, base, offset, count, &count,
1748                                           NULL, NULL);
1749 
1750             if (ret < 0) {
1751                 if (s->salvage) {
1752                     if (n == 1) {
1753                         if (!s->quiet) {
1754                             warn_report("error while reading block status at "
1755                                         "offset %" PRIu64 ": %s", offset,
1756                                         strerror(-ret));
1757                         }
1758                         /* Just try to read the data, then */
1759                         ret = BDRV_BLOCK_DATA;
1760                         count = BDRV_SECTOR_SIZE;
1761                     } else {
1762                         /* Retry on a shorter range */
1763                         n = DIV_ROUND_UP(n, 4);
1764                     }
1765                 } else {
1766                     error_report("error while reading block status at offset "
1767                                  "%" PRIu64 ": %s", offset, strerror(-ret));
1768                     return ret;
1769                 }
1770             }
1771         } while (ret < 0);
1772 
1773         n = DIV_ROUND_UP(count, BDRV_SECTOR_SIZE);
1774 
1775         /*
1776          * Avoid that s->sector_next_status becomes unaligned to the source
1777          * request alignment and/or cluster size to avoid unnecessary read
1778          * cycles.
1779          */
1780         tail = (sector_num - src_cur_offset + n) % s->src_alignment[src_cur];
1781         if (n > tail) {
1782             n -= tail;
1783         }
1784 
1785         if (ret & BDRV_BLOCK_ZERO) {
1786             s->status = post_backing_zero ? BLK_BACKING_FILE : BLK_ZERO;
1787         } else if (ret & BDRV_BLOCK_DATA) {
1788             s->status = BLK_DATA;
1789         } else {
1790             s->status = s->target_has_backing ? BLK_BACKING_FILE : BLK_DATA;
1791         }
1792 
1793         s->sector_next_status = sector_num + n;
1794     }
1795 
1796     n = MIN(n, s->sector_next_status - sector_num);
1797     if (s->status == BLK_DATA) {
1798         n = MIN(n, s->buf_sectors);
1799     }
1800 
1801     /* We need to write complete clusters for compressed images, so if an
1802      * unallocated area is shorter than that, we must consider the whole
1803      * cluster allocated. */
1804     if (s->compressed) {
1805         if (n < s->cluster_sectors) {
1806             n = MIN(s->cluster_sectors, s->total_sectors - sector_num);
1807             s->status = BLK_DATA;
1808         } else {
1809             n = QEMU_ALIGN_DOWN(n, s->cluster_sectors);
1810         }
1811     }
1812 
1813     return n;
1814 }
1815 
1816 static int coroutine_fn convert_co_read(ImgConvertState *s, int64_t sector_num,
1817                                         int nb_sectors, uint8_t *buf)
1818 {
1819     uint64_t single_read_until = 0;
1820     int n, ret;
1821 
1822     assert(nb_sectors <= s->buf_sectors);
1823     while (nb_sectors > 0) {
1824         BlockBackend *blk;
1825         int src_cur;
1826         int64_t bs_sectors, src_cur_offset;
1827         uint64_t offset;
1828 
1829         /* In the case of compression with multiple source files, we can get a
1830          * nb_sectors that spreads into the next part. So we must be able to
1831          * read across multiple BDSes for one convert_read() call. */
1832         convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1833         blk = s->src[src_cur];
1834         bs_sectors = s->src_sectors[src_cur];
1835 
1836         offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
1837 
1838         n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
1839         if (single_read_until > offset) {
1840             n = 1;
1841         }
1842 
1843         ret = blk_co_pread(blk, offset, n << BDRV_SECTOR_BITS, buf, 0);
1844         if (ret < 0) {
1845             if (s->salvage) {
1846                 if (n > 1) {
1847                     single_read_until = offset + (n << BDRV_SECTOR_BITS);
1848                     continue;
1849                 } else {
1850                     if (!s->quiet) {
1851                         warn_report("error while reading offset %" PRIu64
1852                                     ": %s", offset, strerror(-ret));
1853                     }
1854                     memset(buf, 0, BDRV_SECTOR_SIZE);
1855                 }
1856             } else {
1857                 return ret;
1858             }
1859         }
1860 
1861         sector_num += n;
1862         nb_sectors -= n;
1863         buf += n * BDRV_SECTOR_SIZE;
1864     }
1865 
1866     return 0;
1867 }
1868 
1869 
1870 static int coroutine_fn convert_co_write(ImgConvertState *s, int64_t sector_num,
1871                                          int nb_sectors, uint8_t *buf,
1872                                          enum ImgConvertBlockStatus status)
1873 {
1874     int ret;
1875 
1876     while (nb_sectors > 0) {
1877         int n = nb_sectors;
1878         BdrvRequestFlags flags = s->compressed ? BDRV_REQ_WRITE_COMPRESSED : 0;
1879 
1880         switch (status) {
1881         case BLK_BACKING_FILE:
1882             /* If we have a backing file, leave clusters unallocated that are
1883              * unallocated in the source image, so that the backing file is
1884              * visible at the respective offset. */
1885             assert(s->target_has_backing);
1886             break;
1887 
1888         case BLK_DATA:
1889             /* If we're told to keep the target fully allocated (-S 0) or there
1890              * is real non-zero data, we must write it. Otherwise we can treat
1891              * it as zero sectors.
1892              * Compressed clusters need to be written as a whole, so in that
1893              * case we can only save the write if the buffer is completely
1894              * zeroed. */
1895             if (!s->min_sparse ||
1896                 (!s->compressed &&
1897                  is_allocated_sectors_min(buf, n, &n, s->min_sparse,
1898                                           sector_num, s->alignment)) ||
1899                 (s->compressed &&
1900                  !buffer_is_zero(buf, n * BDRV_SECTOR_SIZE)))
1901             {
1902                 ret = blk_co_pwrite(s->target, sector_num << BDRV_SECTOR_BITS,
1903                                     n << BDRV_SECTOR_BITS, buf, flags);
1904                 if (ret < 0) {
1905                     return ret;
1906                 }
1907                 break;
1908             }
1909             /* fall-through */
1910 
1911         case BLK_ZERO:
1912             if (s->has_zero_init) {
1913                 assert(!s->target_has_backing);
1914                 break;
1915             }
1916             ret = blk_co_pwrite_zeroes(s->target,
1917                                        sector_num << BDRV_SECTOR_BITS,
1918                                        n << BDRV_SECTOR_BITS,
1919                                        BDRV_REQ_MAY_UNMAP);
1920             if (ret < 0) {
1921                 return ret;
1922             }
1923             break;
1924         }
1925 
1926         sector_num += n;
1927         nb_sectors -= n;
1928         buf += n * BDRV_SECTOR_SIZE;
1929     }
1930 
1931     return 0;
1932 }
1933 
1934 static int coroutine_fn convert_co_copy_range(ImgConvertState *s, int64_t sector_num,
1935                                               int nb_sectors)
1936 {
1937     int n, ret;
1938 
1939     while (nb_sectors > 0) {
1940         BlockBackend *blk;
1941         int src_cur;
1942         int64_t bs_sectors, src_cur_offset;
1943         int64_t offset;
1944 
1945         convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1946         offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
1947         blk = s->src[src_cur];
1948         bs_sectors = s->src_sectors[src_cur];
1949 
1950         n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
1951 
1952         ret = blk_co_copy_range(blk, offset, s->target,
1953                                 sector_num << BDRV_SECTOR_BITS,
1954                                 n << BDRV_SECTOR_BITS, 0, 0);
1955         if (ret < 0) {
1956             return ret;
1957         }
1958 
1959         sector_num += n;
1960         nb_sectors -= n;
1961     }
1962     return 0;
1963 }
1964 
1965 static void coroutine_fn convert_co_do_copy(void *opaque)
1966 {
1967     ImgConvertState *s = opaque;
1968     uint8_t *buf = NULL;
1969     int ret, i;
1970     int index = -1;
1971 
1972     for (i = 0; i < s->num_coroutines; i++) {
1973         if (s->co[i] == qemu_coroutine_self()) {
1974             index = i;
1975             break;
1976         }
1977     }
1978     assert(index >= 0);
1979 
1980     s->running_coroutines++;
1981     buf = blk_blockalign(s->target, s->buf_sectors * BDRV_SECTOR_SIZE);
1982 
1983     while (1) {
1984         int n;
1985         int64_t sector_num;
1986         enum ImgConvertBlockStatus status;
1987         bool copy_range;
1988 
1989         qemu_co_mutex_lock(&s->lock);
1990         if (s->ret != -EINPROGRESS || s->sector_num >= s->total_sectors) {
1991             qemu_co_mutex_unlock(&s->lock);
1992             break;
1993         }
1994         WITH_GRAPH_RDLOCK_GUARD() {
1995             n = convert_iteration_sectors(s, s->sector_num);
1996         }
1997         if (n < 0) {
1998             qemu_co_mutex_unlock(&s->lock);
1999             s->ret = n;
2000             break;
2001         }
2002         /* save current sector and allocation status to local variables */
2003         sector_num = s->sector_num;
2004         status = s->status;
2005         if (!s->min_sparse && s->status == BLK_ZERO) {
2006             n = MIN(n, s->buf_sectors);
2007         }
2008         /* increment global sector counter so that other coroutines can
2009          * already continue reading beyond this request */
2010         s->sector_num += n;
2011         qemu_co_mutex_unlock(&s->lock);
2012 
2013         if (status == BLK_DATA || (!s->min_sparse && status == BLK_ZERO)) {
2014             s->allocated_done += n;
2015             qemu_progress_print(100.0 * s->allocated_done /
2016                                         s->allocated_sectors, 0);
2017         }
2018 
2019 retry:
2020         copy_range = s->copy_range && s->status == BLK_DATA;
2021         if (status == BLK_DATA && !copy_range) {
2022             ret = convert_co_read(s, sector_num, n, buf);
2023             if (ret < 0) {
2024                 error_report("error while reading at byte %lld: %s",
2025                              sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
2026                 s->ret = ret;
2027             }
2028         } else if (!s->min_sparse && status == BLK_ZERO) {
2029             status = BLK_DATA;
2030             memset(buf, 0x00, n * BDRV_SECTOR_SIZE);
2031         }
2032 
2033         if (s->wr_in_order) {
2034             /* keep writes in order */
2035             while (s->wr_offs != sector_num && s->ret == -EINPROGRESS) {
2036                 s->wait_sector_num[index] = sector_num;
2037                 qemu_coroutine_yield();
2038             }
2039             s->wait_sector_num[index] = -1;
2040         }
2041 
2042         if (s->ret == -EINPROGRESS) {
2043             if (copy_range) {
2044                 WITH_GRAPH_RDLOCK_GUARD() {
2045                     ret = convert_co_copy_range(s, sector_num, n);
2046                 }
2047                 if (ret) {
2048                     s->copy_range = false;
2049                     goto retry;
2050                 }
2051             } else {
2052                 ret = convert_co_write(s, sector_num, n, buf, status);
2053             }
2054             if (ret < 0) {
2055                 error_report("error while writing at byte %lld: %s",
2056                              sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
2057                 s->ret = ret;
2058             }
2059         }
2060 
2061         if (s->wr_in_order) {
2062             /* reenter the coroutine that might have waited
2063              * for this write to complete */
2064             s->wr_offs = sector_num + n;
2065             for (i = 0; i < s->num_coroutines; i++) {
2066                 if (s->co[i] && s->wait_sector_num[i] == s->wr_offs) {
2067                     /*
2068                      * A -> B -> A cannot occur because A has
2069                      * s->wait_sector_num[i] == -1 during A -> B.  Therefore
2070                      * B will never enter A during this time window.
2071                      */
2072                     qemu_coroutine_enter(s->co[i]);
2073                     break;
2074                 }
2075             }
2076         }
2077     }
2078 
2079     qemu_vfree(buf);
2080     s->co[index] = NULL;
2081     s->running_coroutines--;
2082     if (!s->running_coroutines && s->ret == -EINPROGRESS) {
2083         /* the convert job finished successfully */
2084         s->ret = 0;
2085     }
2086 }
2087 
2088 static int convert_do_copy(ImgConvertState *s)
2089 {
2090     int ret, i, n;
2091     int64_t sector_num = 0;
2092 
2093     /* Check whether we have zero initialisation or can get it efficiently */
2094     if (!s->has_zero_init && s->target_is_new && s->min_sparse &&
2095         !s->target_has_backing) {
2096         s->has_zero_init = bdrv_has_zero_init(blk_bs(s->target));
2097     }
2098 
2099     /* Allocate buffer for copied data. For compressed images, only one cluster
2100      * can be copied at a time. */
2101     if (s->compressed) {
2102         if (s->cluster_sectors <= 0 || s->cluster_sectors > s->buf_sectors) {
2103             error_report("invalid cluster size");
2104             return -EINVAL;
2105         }
2106         s->buf_sectors = s->cluster_sectors;
2107     }
2108 
2109     while (sector_num < s->total_sectors) {
2110         n = convert_iteration_sectors(s, sector_num);
2111         if (n < 0) {
2112             return n;
2113         }
2114         if (s->status == BLK_DATA || (!s->min_sparse && s->status == BLK_ZERO))
2115         {
2116             s->allocated_sectors += n;
2117         }
2118         sector_num += n;
2119     }
2120 
2121     /* Do the copy */
2122     s->sector_next_status = 0;
2123     s->ret = -EINPROGRESS;
2124 
2125     qemu_co_mutex_init(&s->lock);
2126     for (i = 0; i < s->num_coroutines; i++) {
2127         s->co[i] = qemu_coroutine_create(convert_co_do_copy, s);
2128         s->wait_sector_num[i] = -1;
2129         qemu_coroutine_enter(s->co[i]);
2130     }
2131 
2132     while (s->running_coroutines) {
2133         main_loop_wait(false);
2134     }
2135 
2136     if (s->compressed && !s->ret) {
2137         /* signal EOF to align */
2138         ret = blk_pwrite_compressed(s->target, 0, 0, NULL);
2139         if (ret < 0) {
2140             return ret;
2141         }
2142     }
2143 
2144     return s->ret;
2145 }
2146 
2147 /* Check that bitmaps can be copied, or output an error */
2148 static int convert_check_bitmaps(BlockDriverState *src, bool skip_broken)
2149 {
2150     BdrvDirtyBitmap *bm;
2151 
2152     if (!bdrv_supports_persistent_dirty_bitmap(src)) {
2153         error_report("Source lacks bitmap support");
2154         return -1;
2155     }
2156     FOR_EACH_DIRTY_BITMAP(src, bm) {
2157         if (!bdrv_dirty_bitmap_get_persistence(bm)) {
2158             continue;
2159         }
2160         if (!skip_broken && bdrv_dirty_bitmap_inconsistent(bm)) {
2161             error_report("Cannot copy inconsistent bitmap '%s'",
2162                          bdrv_dirty_bitmap_name(bm));
2163             error_printf("Try --skip-broken-bitmaps, or "
2164                          "use 'qemu-img bitmap --remove' to delete it\n");
2165             return -1;
2166         }
2167     }
2168     return 0;
2169 }
2170 
2171 static int convert_copy_bitmaps(BlockDriverState *src, BlockDriverState *dst,
2172                                 bool skip_broken)
2173 {
2174     BdrvDirtyBitmap *bm;
2175     Error *err = NULL;
2176 
2177     FOR_EACH_DIRTY_BITMAP(src, bm) {
2178         const char *name;
2179 
2180         if (!bdrv_dirty_bitmap_get_persistence(bm)) {
2181             continue;
2182         }
2183         name = bdrv_dirty_bitmap_name(bm);
2184         if (skip_broken && bdrv_dirty_bitmap_inconsistent(bm)) {
2185             warn_report("Skipping inconsistent bitmap '%s'", name);
2186             continue;
2187         }
2188         qmp_block_dirty_bitmap_add(dst->node_name, name,
2189                                    true, bdrv_dirty_bitmap_granularity(bm),
2190                                    true, true,
2191                                    true, !bdrv_dirty_bitmap_enabled(bm),
2192                                    &err);
2193         if (err) {
2194             error_reportf_err(err, "Failed to create bitmap %s: ", name);
2195             return -1;
2196         }
2197 
2198         do_dirty_bitmap_merge(dst->node_name, name, src->node_name, name,
2199                               &err);
2200         if (err) {
2201             error_reportf_err(err, "Failed to populate bitmap %s: ", name);
2202             qmp_block_dirty_bitmap_remove(dst->node_name, name, NULL);
2203             return -1;
2204         }
2205     }
2206 
2207     return 0;
2208 }
2209 
2210 #define MAX_BUF_SECTORS 32768
2211 
2212 static void set_rate_limit(BlockBackend *blk, int64_t rate_limit)
2213 {
2214     ThrottleConfig cfg;
2215 
2216     throttle_config_init(&cfg);
2217     cfg.buckets[THROTTLE_BPS_WRITE].avg = rate_limit;
2218 
2219     blk_io_limits_enable(blk, CONVERT_THROTTLE_GROUP);
2220     blk_set_io_limits(blk, &cfg);
2221 }
2222 
2223 static int img_convert(int argc, char **argv)
2224 {
2225     int c, bs_i, flags, src_flags = BDRV_O_NO_SHARE;
2226     const char *fmt = NULL, *out_fmt = NULL, *cache = "unsafe",
2227                *src_cache = BDRV_DEFAULT_CACHE, *out_baseimg = NULL,
2228                *out_filename, *out_baseimg_param, *snapshot_name = NULL,
2229                *backing_fmt = NULL;
2230     BlockDriver *drv = NULL, *proto_drv = NULL;
2231     BlockDriverInfo bdi;
2232     BlockDriverState *out_bs;
2233     QemuOpts *opts = NULL, *sn_opts = NULL;
2234     QemuOptsList *create_opts = NULL;
2235     QDict *open_opts = NULL;
2236     char *options = NULL;
2237     Error *local_err = NULL;
2238     bool writethrough, src_writethrough, image_opts = false,
2239          skip_create = false, progress = false, tgt_image_opts = false;
2240     int64_t ret = -EINVAL;
2241     bool force_share = false;
2242     bool explict_min_sparse = false;
2243     bool bitmaps = false;
2244     bool skip_broken = false;
2245     int64_t rate_limit = 0;
2246 
2247     ImgConvertState s = (ImgConvertState) {
2248         /* Need at least 4k of zeros for sparse detection */
2249         .min_sparse         = 8,
2250         .copy_range         = false,
2251         .buf_sectors        = IO_BUF_SIZE / BDRV_SECTOR_SIZE,
2252         .wr_in_order        = true,
2253         .num_coroutines     = 8,
2254     };
2255 
2256     for(;;) {
2257         static const struct option long_options[] = {
2258             {"help", no_argument, 0, 'h'},
2259             {"object", required_argument, 0, OPTION_OBJECT},
2260             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2261             {"force-share", no_argument, 0, 'U'},
2262             {"target-image-opts", no_argument, 0, OPTION_TARGET_IMAGE_OPTS},
2263             {"salvage", no_argument, 0, OPTION_SALVAGE},
2264             {"target-is-zero", no_argument, 0, OPTION_TARGET_IS_ZERO},
2265             {"bitmaps", no_argument, 0, OPTION_BITMAPS},
2266             {"skip-broken-bitmaps", no_argument, 0, OPTION_SKIP_BROKEN},
2267             {0, 0, 0, 0}
2268         };
2269         c = getopt_long(argc, argv, ":hf:O:B:CcF:o:l:S:pt:T:qnm:WUr:",
2270                         long_options, NULL);
2271         if (c == -1) {
2272             break;
2273         }
2274         switch(c) {
2275         case ':':
2276             missing_argument(argv[optind - 1]);
2277             break;
2278         case '?':
2279             unrecognized_option(argv[optind - 1]);
2280             break;
2281         case 'h':
2282             help();
2283             break;
2284         case 'f':
2285             fmt = optarg;
2286             break;
2287         case 'O':
2288             out_fmt = optarg;
2289             break;
2290         case 'B':
2291             out_baseimg = optarg;
2292             break;
2293         case 'C':
2294             s.copy_range = true;
2295             break;
2296         case 'c':
2297             s.compressed = true;
2298             break;
2299         case 'F':
2300             backing_fmt = optarg;
2301             break;
2302         case 'o':
2303             if (accumulate_options(&options, optarg) < 0) {
2304                 goto fail_getopt;
2305             }
2306             break;
2307         case 'l':
2308             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
2309                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
2310                                                   optarg, false);
2311                 if (!sn_opts) {
2312                     error_report("Failed in parsing snapshot param '%s'",
2313                                  optarg);
2314                     goto fail_getopt;
2315                 }
2316             } else {
2317                 snapshot_name = optarg;
2318             }
2319             break;
2320         case 'S':
2321         {
2322             int64_t sval;
2323 
2324             sval = cvtnum("buffer size for sparse output", optarg);
2325             if (sval < 0) {
2326                 goto fail_getopt;
2327             } else if (!QEMU_IS_ALIGNED(sval, BDRV_SECTOR_SIZE) ||
2328                 sval / BDRV_SECTOR_SIZE > MAX_BUF_SECTORS) {
2329                 error_report("Invalid buffer size for sparse output specified. "
2330                     "Valid sizes are multiples of %llu up to %llu. Select "
2331                     "0 to disable sparse detection (fully allocates output).",
2332                     BDRV_SECTOR_SIZE, MAX_BUF_SECTORS * BDRV_SECTOR_SIZE);
2333                 goto fail_getopt;
2334             }
2335 
2336             s.min_sparse = sval / BDRV_SECTOR_SIZE;
2337             explict_min_sparse = true;
2338             break;
2339         }
2340         case 'p':
2341             progress = true;
2342             break;
2343         case 't':
2344             cache = optarg;
2345             break;
2346         case 'T':
2347             src_cache = optarg;
2348             break;
2349         case 'q':
2350             s.quiet = true;
2351             break;
2352         case 'n':
2353             skip_create = true;
2354             break;
2355         case 'm':
2356             if (qemu_strtol(optarg, NULL, 0, &s.num_coroutines) ||
2357                 s.num_coroutines < 1 || s.num_coroutines > MAX_COROUTINES) {
2358                 error_report("Invalid number of coroutines. Allowed number of"
2359                              " coroutines is between 1 and %d", MAX_COROUTINES);
2360                 goto fail_getopt;
2361             }
2362             break;
2363         case 'W':
2364             s.wr_in_order = false;
2365             break;
2366         case 'U':
2367             force_share = true;
2368             break;
2369         case 'r':
2370             rate_limit = cvtnum("rate limit", optarg);
2371             if (rate_limit < 0) {
2372                 goto fail_getopt;
2373             }
2374             break;
2375         case OPTION_OBJECT:
2376             user_creatable_process_cmdline(optarg);
2377             break;
2378         case OPTION_IMAGE_OPTS:
2379             image_opts = true;
2380             break;
2381         case OPTION_SALVAGE:
2382             s.salvage = true;
2383             break;
2384         case OPTION_TARGET_IMAGE_OPTS:
2385             tgt_image_opts = true;
2386             break;
2387         case OPTION_TARGET_IS_ZERO:
2388             /*
2389              * The user asserting that the target is blank has the
2390              * same effect as the target driver supporting zero
2391              * initialisation.
2392              */
2393             s.has_zero_init = true;
2394             break;
2395         case OPTION_BITMAPS:
2396             bitmaps = true;
2397             break;
2398         case OPTION_SKIP_BROKEN:
2399             skip_broken = true;
2400             break;
2401         }
2402     }
2403 
2404     if (!out_fmt && !tgt_image_opts) {
2405         out_fmt = "raw";
2406     }
2407 
2408     if (skip_broken && !bitmaps) {
2409         error_report("Use of --skip-broken-bitmaps requires --bitmaps");
2410         goto fail_getopt;
2411     }
2412 
2413     if (s.compressed && s.copy_range) {
2414         error_report("Cannot enable copy offloading when -c is used");
2415         goto fail_getopt;
2416     }
2417 
2418     if (explict_min_sparse && s.copy_range) {
2419         error_report("Cannot enable copy offloading when -S is used");
2420         goto fail_getopt;
2421     }
2422 
2423     if (s.copy_range && s.salvage) {
2424         error_report("Cannot use copy offloading in salvaging mode");
2425         goto fail_getopt;
2426     }
2427 
2428     if (tgt_image_opts && !skip_create) {
2429         error_report("--target-image-opts requires use of -n flag");
2430         goto fail_getopt;
2431     }
2432 
2433     if (skip_create && options) {
2434         error_report("-o has no effect when skipping image creation");
2435         goto fail_getopt;
2436     }
2437 
2438     if (s.has_zero_init && !skip_create) {
2439         error_report("--target-is-zero requires use of -n flag");
2440         goto fail_getopt;
2441     }
2442 
2443     s.src_num = argc - optind - 1;
2444     out_filename = s.src_num >= 1 ? argv[argc - 1] : NULL;
2445 
2446     if (options && has_help_option(options)) {
2447         if (out_fmt) {
2448             ret = print_block_option_help(out_filename, out_fmt);
2449             goto fail_getopt;
2450         } else {
2451             error_report("Option help requires a format be specified");
2452             goto fail_getopt;
2453         }
2454     }
2455 
2456     if (s.src_num < 1) {
2457         error_report("Must specify image file name");
2458         goto fail_getopt;
2459     }
2460 
2461     /* ret is still -EINVAL until here */
2462     ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
2463     if (ret < 0) {
2464         error_report("Invalid source cache option: %s", src_cache);
2465         goto fail_getopt;
2466     }
2467 
2468     /* Initialize before goto out */
2469     if (s.quiet) {
2470         progress = false;
2471     }
2472     qemu_progress_init(progress, 1.0);
2473     qemu_progress_print(0, 100);
2474 
2475     s.src = g_new0(BlockBackend *, s.src_num);
2476     s.src_sectors = g_new(int64_t, s.src_num);
2477     s.src_alignment = g_new(int, s.src_num);
2478 
2479     for (bs_i = 0; bs_i < s.src_num; bs_i++) {
2480         BlockDriverState *src_bs;
2481         s.src[bs_i] = img_open(image_opts, argv[optind + bs_i],
2482                                fmt, src_flags, src_writethrough, s.quiet,
2483                                force_share);
2484         if (!s.src[bs_i]) {
2485             ret = -1;
2486             goto out;
2487         }
2488         s.src_sectors[bs_i] = blk_nb_sectors(s.src[bs_i]);
2489         if (s.src_sectors[bs_i] < 0) {
2490             error_report("Could not get size of %s: %s",
2491                          argv[optind + bs_i], strerror(-s.src_sectors[bs_i]));
2492             ret = -1;
2493             goto out;
2494         }
2495         src_bs = blk_bs(s.src[bs_i]);
2496         s.src_alignment[bs_i] = DIV_ROUND_UP(src_bs->bl.request_alignment,
2497                                              BDRV_SECTOR_SIZE);
2498         if (!bdrv_get_info(src_bs, &bdi)) {
2499             s.src_alignment[bs_i] = MAX(s.src_alignment[bs_i],
2500                                         bdi.cluster_size / BDRV_SECTOR_SIZE);
2501         }
2502         s.total_sectors += s.src_sectors[bs_i];
2503     }
2504 
2505     if (sn_opts) {
2506         bdrv_snapshot_load_tmp(blk_bs(s.src[0]),
2507                                qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
2508                                qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
2509                                &local_err);
2510     } else if (snapshot_name != NULL) {
2511         if (s.src_num > 1) {
2512             error_report("No support for concatenating multiple snapshot");
2513             ret = -1;
2514             goto out;
2515         }
2516 
2517         bdrv_snapshot_load_tmp_by_id_or_name(blk_bs(s.src[0]), snapshot_name,
2518                                              &local_err);
2519     }
2520     if (local_err) {
2521         error_reportf_err(local_err, "Failed to load snapshot: ");
2522         ret = -1;
2523         goto out;
2524     }
2525 
2526     if (!skip_create) {
2527         /* Find driver and parse its options */
2528         drv = bdrv_find_format(out_fmt);
2529         if (!drv) {
2530             error_report("Unknown file format '%s'", out_fmt);
2531             ret = -1;
2532             goto out;
2533         }
2534 
2535         proto_drv = bdrv_find_protocol(out_filename, true, &local_err);
2536         if (!proto_drv) {
2537             error_report_err(local_err);
2538             ret = -1;
2539             goto out;
2540         }
2541 
2542         if (!drv->create_opts) {
2543             error_report("Format driver '%s' does not support image creation",
2544                          drv->format_name);
2545             ret = -1;
2546             goto out;
2547         }
2548 
2549         if (!proto_drv->create_opts) {
2550             error_report("Protocol driver '%s' does not support image creation",
2551                          proto_drv->format_name);
2552             ret = -1;
2553             goto out;
2554         }
2555 
2556         create_opts = qemu_opts_append(create_opts, drv->create_opts);
2557         create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
2558 
2559         opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
2560         if (options) {
2561             if (!qemu_opts_do_parse(opts, options, NULL, &local_err)) {
2562                 error_report_err(local_err);
2563                 ret = -1;
2564                 goto out;
2565             }
2566         }
2567 
2568         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2569                             s.total_sectors * BDRV_SECTOR_SIZE, &error_abort);
2570         ret = add_old_style_options(out_fmt, opts, out_baseimg, backing_fmt);
2571         if (ret < 0) {
2572             goto out;
2573         }
2574     }
2575 
2576     /* Get backing file name if -o backing_file was used */
2577     out_baseimg_param = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
2578     if (out_baseimg_param) {
2579         out_baseimg = out_baseimg_param;
2580     }
2581     s.target_has_backing = (bool) out_baseimg;
2582 
2583     if (s.has_zero_init && s.target_has_backing) {
2584         error_report("Cannot use --target-is-zero when the destination "
2585                      "image has a backing file");
2586         goto out;
2587     }
2588 
2589     if (s.src_num > 1 && out_baseimg) {
2590         error_report("Having a backing file for the target makes no sense when "
2591                      "concatenating multiple input images");
2592         ret = -1;
2593         goto out;
2594     }
2595 
2596     if (out_baseimg_param) {
2597         if (!qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT)) {
2598             error_report("Use of backing file requires explicit "
2599                          "backing format");
2600             ret = -1;
2601             goto out;
2602         }
2603     }
2604 
2605     /* Check if compression is supported */
2606     if (s.compressed) {
2607         bool encryption =
2608             qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, false);
2609         const char *encryptfmt =
2610             qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT);
2611         const char *preallocation =
2612             qemu_opt_get(opts, BLOCK_OPT_PREALLOC);
2613 
2614         if (drv && !block_driver_can_compress(drv)) {
2615             error_report("Compression not supported for this file format");
2616             ret = -1;
2617             goto out;
2618         }
2619 
2620         if (encryption || encryptfmt) {
2621             error_report("Compression and encryption not supported at "
2622                          "the same time");
2623             ret = -1;
2624             goto out;
2625         }
2626 
2627         if (preallocation
2628             && strcmp(preallocation, "off"))
2629         {
2630             error_report("Compression and preallocation not supported at "
2631                          "the same time");
2632             ret = -1;
2633             goto out;
2634         }
2635     }
2636 
2637     /* Determine if bitmaps need copying */
2638     if (bitmaps) {
2639         if (s.src_num > 1) {
2640             error_report("Copying bitmaps only possible with single source");
2641             ret = -1;
2642             goto out;
2643         }
2644         ret = convert_check_bitmaps(blk_bs(s.src[0]), skip_broken);
2645         if (ret < 0) {
2646             goto out;
2647         }
2648     }
2649 
2650     /*
2651      * The later open call will need any decryption secrets, and
2652      * bdrv_create() will purge "opts", so extract them now before
2653      * they are lost.
2654      */
2655     if (!skip_create) {
2656         open_opts = qdict_new();
2657         qemu_opt_foreach(opts, img_add_key_secrets, open_opts, &error_abort);
2658 
2659         /* Create the new image */
2660         ret = bdrv_create(drv, out_filename, opts, &local_err);
2661         if (ret < 0) {
2662             error_reportf_err(local_err, "%s: error while converting %s: ",
2663                               out_filename, out_fmt);
2664             goto out;
2665         }
2666     }
2667 
2668     s.target_is_new = !skip_create;
2669 
2670     flags = s.min_sparse ? (BDRV_O_RDWR | BDRV_O_UNMAP) : BDRV_O_RDWR;
2671     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
2672     if (ret < 0) {
2673         error_report("Invalid cache option: %s", cache);
2674         goto out;
2675     }
2676 
2677     if (flags & BDRV_O_NOCACHE) {
2678         /*
2679          * If we open the target with O_DIRECT, it may be necessary to
2680          * extend its size to align to the physical sector size.
2681          */
2682         flags |= BDRV_O_RESIZE;
2683     }
2684 
2685     if (skip_create) {
2686         s.target = img_open(tgt_image_opts, out_filename, out_fmt,
2687                             flags, writethrough, s.quiet, false);
2688     } else {
2689         /* TODO ultimately we should allow --target-image-opts
2690          * to be used even when -n is not given.
2691          * That has to wait for bdrv_create to be improved
2692          * to allow filenames in option syntax
2693          */
2694         s.target = img_open_file(out_filename, open_opts, out_fmt,
2695                                  flags, writethrough, s.quiet, false);
2696         open_opts = NULL; /* blk_new_open will have freed it */
2697     }
2698     if (!s.target) {
2699         ret = -1;
2700         goto out;
2701     }
2702     out_bs = blk_bs(s.target);
2703 
2704     if (bitmaps && !bdrv_supports_persistent_dirty_bitmap(out_bs)) {
2705         error_report("Format driver '%s' does not support bitmaps",
2706                      out_bs->drv->format_name);
2707         ret = -1;
2708         goto out;
2709     }
2710 
2711     if (s.compressed && !block_driver_can_compress(out_bs->drv)) {
2712         error_report("Compression not supported for this file format");
2713         ret = -1;
2714         goto out;
2715     }
2716 
2717     /* increase bufsectors from the default 4096 (2M) if opt_transfer
2718      * or discard_alignment of the out_bs is greater. Limit to
2719      * MAX_BUF_SECTORS as maximum which is currently 32768 (16MB). */
2720     s.buf_sectors = MIN(MAX_BUF_SECTORS,
2721                         MAX(s.buf_sectors,
2722                             MAX(out_bs->bl.opt_transfer >> BDRV_SECTOR_BITS,
2723                                 out_bs->bl.pdiscard_alignment >>
2724                                 BDRV_SECTOR_BITS)));
2725 
2726     /* try to align the write requests to the destination to avoid unnecessary
2727      * RMW cycles. */
2728     s.alignment = MAX(pow2floor(s.min_sparse),
2729                       DIV_ROUND_UP(out_bs->bl.request_alignment,
2730                                    BDRV_SECTOR_SIZE));
2731     assert(is_power_of_2(s.alignment));
2732 
2733     if (skip_create) {
2734         int64_t output_sectors = blk_nb_sectors(s.target);
2735         if (output_sectors < 0) {
2736             error_report("unable to get output image length: %s",
2737                          strerror(-output_sectors));
2738             ret = -1;
2739             goto out;
2740         } else if (output_sectors < s.total_sectors) {
2741             error_report("output file is smaller than input file");
2742             ret = -1;
2743             goto out;
2744         }
2745     }
2746 
2747     if (s.target_has_backing && s.target_is_new) {
2748         /* Errors are treated as "backing length unknown" (which means
2749          * s.target_backing_sectors has to be negative, which it will
2750          * be automatically).  The backing file length is used only
2751          * for optimizations, so such a case is not fatal. */
2752         s.target_backing_sectors =
2753             bdrv_nb_sectors(bdrv_backing_chain_next(out_bs));
2754     } else {
2755         s.target_backing_sectors = -1;
2756     }
2757 
2758     ret = bdrv_get_info(out_bs, &bdi);
2759     if (ret < 0) {
2760         if (s.compressed) {
2761             error_report("could not get block driver info");
2762             goto out;
2763         }
2764     } else {
2765         s.compressed = s.compressed || bdi.needs_compressed_writes;
2766         s.cluster_sectors = bdi.cluster_size / BDRV_SECTOR_SIZE;
2767     }
2768 
2769     if (rate_limit) {
2770         set_rate_limit(s.target, rate_limit);
2771     }
2772 
2773     ret = convert_do_copy(&s);
2774 
2775     /* Now copy the bitmaps */
2776     if (bitmaps && ret == 0) {
2777         ret = convert_copy_bitmaps(blk_bs(s.src[0]), out_bs, skip_broken);
2778     }
2779 
2780 out:
2781     if (!ret) {
2782         qemu_progress_print(100, 0);
2783     }
2784     qemu_progress_end();
2785     qemu_opts_del(opts);
2786     qemu_opts_free(create_opts);
2787     qobject_unref(open_opts);
2788     blk_unref(s.target);
2789     if (s.src) {
2790         for (bs_i = 0; bs_i < s.src_num; bs_i++) {
2791             blk_unref(s.src[bs_i]);
2792         }
2793         g_free(s.src);
2794     }
2795     g_free(s.src_sectors);
2796     g_free(s.src_alignment);
2797 fail_getopt:
2798     qemu_opts_del(sn_opts);
2799     g_free(options);
2800 
2801     return !!ret;
2802 }
2803 
2804 
2805 static void dump_snapshots(BlockDriverState *bs)
2806 {
2807     QEMUSnapshotInfo *sn_tab, *sn;
2808     int nb_sns, i;
2809 
2810     nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2811     if (nb_sns <= 0)
2812         return;
2813     printf("Snapshot list:\n");
2814     bdrv_snapshot_dump(NULL);
2815     printf("\n");
2816     for(i = 0; i < nb_sns; i++) {
2817         sn = &sn_tab[i];
2818         bdrv_snapshot_dump(sn);
2819         printf("\n");
2820     }
2821     g_free(sn_tab);
2822 }
2823 
2824 static void dump_json_block_graph_info_list(BlockGraphInfoList *list)
2825 {
2826     GString *str;
2827     QObject *obj;
2828     Visitor *v = qobject_output_visitor_new(&obj);
2829 
2830     visit_type_BlockGraphInfoList(v, NULL, &list, &error_abort);
2831     visit_complete(v, &obj);
2832     str = qobject_to_json_pretty(obj, true);
2833     assert(str != NULL);
2834     printf("%s\n", str->str);
2835     qobject_unref(obj);
2836     visit_free(v);
2837     g_string_free(str, true);
2838 }
2839 
2840 static void dump_json_block_graph_info(BlockGraphInfo *info)
2841 {
2842     GString *str;
2843     QObject *obj;
2844     Visitor *v = qobject_output_visitor_new(&obj);
2845 
2846     visit_type_BlockGraphInfo(v, NULL, &info, &error_abort);
2847     visit_complete(v, &obj);
2848     str = qobject_to_json_pretty(obj, true);
2849     assert(str != NULL);
2850     printf("%s\n", str->str);
2851     qobject_unref(obj);
2852     visit_free(v);
2853     g_string_free(str, true);
2854 }
2855 
2856 static void dump_human_image_info(BlockGraphInfo *info, int indentation,
2857                                   const char *path)
2858 {
2859     BlockChildInfoList *children_list;
2860 
2861     bdrv_node_info_dump(qapi_BlockGraphInfo_base(info), indentation,
2862                         info->children == NULL);
2863 
2864     for (children_list = info->children; children_list;
2865          children_list = children_list->next)
2866     {
2867         BlockChildInfo *child = children_list->value;
2868         g_autofree char *child_path = NULL;
2869 
2870         printf("%*sChild node '%s%s':\n",
2871                indentation * 4, "", path, child->name);
2872         child_path = g_strdup_printf("%s%s/", path, child->name);
2873         dump_human_image_info(child->info, indentation + 1, child_path);
2874     }
2875 }
2876 
2877 static void dump_human_image_info_list(BlockGraphInfoList *list)
2878 {
2879     BlockGraphInfoList *elem;
2880     bool delim = false;
2881 
2882     for (elem = list; elem; elem = elem->next) {
2883         if (delim) {
2884             printf("\n");
2885         }
2886         delim = true;
2887 
2888         dump_human_image_info(elem->value, 0, "/");
2889     }
2890 }
2891 
2892 static gboolean str_equal_func(gconstpointer a, gconstpointer b)
2893 {
2894     return strcmp(a, b) == 0;
2895 }
2896 
2897 /**
2898  * Open an image file chain and return an BlockGraphInfoList
2899  *
2900  * @filename: topmost image filename
2901  * @fmt: topmost image format (may be NULL to autodetect)
2902  * @chain: true  - enumerate entire backing file chain
2903  *         false - only topmost image file
2904  *
2905  * Returns a list of BlockNodeInfo objects or NULL if there was an error
2906  * opening an image file.  If there was an error a message will have been
2907  * printed to stderr.
2908  */
2909 static BlockGraphInfoList *collect_image_info_list(bool image_opts,
2910                                                    const char *filename,
2911                                                    const char *fmt,
2912                                                    bool chain, bool force_share)
2913 {
2914     BlockGraphInfoList *head = NULL;
2915     BlockGraphInfoList **tail = &head;
2916     GHashTable *filenames;
2917     Error *err = NULL;
2918 
2919     filenames = g_hash_table_new_full(g_str_hash, str_equal_func, NULL, NULL);
2920 
2921     while (filename) {
2922         BlockBackend *blk;
2923         BlockDriverState *bs;
2924         BlockGraphInfo *info;
2925 
2926         if (g_hash_table_lookup_extended(filenames, filename, NULL, NULL)) {
2927             error_report("Backing file '%s' creates an infinite loop.",
2928                          filename);
2929             goto err;
2930         }
2931         g_hash_table_insert(filenames, (gpointer)filename, NULL);
2932 
2933         blk = img_open(image_opts, filename, fmt,
2934                        BDRV_O_NO_BACKING | BDRV_O_NO_IO, false, false,
2935                        force_share);
2936         if (!blk) {
2937             goto err;
2938         }
2939         bs = blk_bs(blk);
2940 
2941         /*
2942          * Note that the returned BlockGraphInfo object will not have
2943          * information about this image's backing node, because we have opened
2944          * it with BDRV_O_NO_BACKING.  Printing this object will therefore not
2945          * duplicate the backing chain information that we obtain by walking
2946          * the chain manually here.
2947          */
2948         bdrv_query_block_graph_info(bs, &info, &err);
2949         if (err) {
2950             error_report_err(err);
2951             blk_unref(blk);
2952             goto err;
2953         }
2954 
2955         QAPI_LIST_APPEND(tail, info);
2956 
2957         blk_unref(blk);
2958 
2959         /* Clear parameters that only apply to the topmost image */
2960         filename = fmt = NULL;
2961         image_opts = false;
2962 
2963         if (chain) {
2964             if (info->full_backing_filename) {
2965                 filename = info->full_backing_filename;
2966             } else if (info->backing_filename) {
2967                 error_report("Could not determine absolute backing filename,"
2968                              " but backing filename '%s' present",
2969                              info->backing_filename);
2970                 goto err;
2971             }
2972             if (info->backing_filename_format) {
2973                 fmt = info->backing_filename_format;
2974             }
2975         }
2976     }
2977     g_hash_table_destroy(filenames);
2978     return head;
2979 
2980 err:
2981     qapi_free_BlockGraphInfoList(head);
2982     g_hash_table_destroy(filenames);
2983     return NULL;
2984 }
2985 
2986 static int img_info(int argc, char **argv)
2987 {
2988     int c;
2989     OutputFormat output_format = OFORMAT_HUMAN;
2990     bool chain = false;
2991     const char *filename, *fmt, *output;
2992     BlockGraphInfoList *list;
2993     bool image_opts = false;
2994     bool force_share = false;
2995 
2996     fmt = NULL;
2997     output = NULL;
2998     for(;;) {
2999         int option_index = 0;
3000         static const struct option long_options[] = {
3001             {"help", no_argument, 0, 'h'},
3002             {"format", required_argument, 0, 'f'},
3003             {"output", required_argument, 0, OPTION_OUTPUT},
3004             {"backing-chain", no_argument, 0, OPTION_BACKING_CHAIN},
3005             {"object", required_argument, 0, OPTION_OBJECT},
3006             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3007             {"force-share", no_argument, 0, 'U'},
3008             {0, 0, 0, 0}
3009         };
3010         c = getopt_long(argc, argv, ":f:hU",
3011                         long_options, &option_index);
3012         if (c == -1) {
3013             break;
3014         }
3015         switch(c) {
3016         case ':':
3017             missing_argument(argv[optind - 1]);
3018             break;
3019         case '?':
3020             unrecognized_option(argv[optind - 1]);
3021             break;
3022         case 'h':
3023             help();
3024             break;
3025         case 'f':
3026             fmt = optarg;
3027             break;
3028         case 'U':
3029             force_share = true;
3030             break;
3031         case OPTION_OUTPUT:
3032             output = optarg;
3033             break;
3034         case OPTION_BACKING_CHAIN:
3035             chain = true;
3036             break;
3037         case OPTION_OBJECT:
3038             user_creatable_process_cmdline(optarg);
3039             break;
3040         case OPTION_IMAGE_OPTS:
3041             image_opts = true;
3042             break;
3043         }
3044     }
3045     if (optind != argc - 1) {
3046         error_exit("Expecting one image file name");
3047     }
3048     filename = argv[optind++];
3049 
3050     if (output && !strcmp(output, "json")) {
3051         output_format = OFORMAT_JSON;
3052     } else if (output && !strcmp(output, "human")) {
3053         output_format = OFORMAT_HUMAN;
3054     } else if (output) {
3055         error_report("--output must be used with human or json as argument.");
3056         return 1;
3057     }
3058 
3059     list = collect_image_info_list(image_opts, filename, fmt, chain,
3060                                    force_share);
3061     if (!list) {
3062         return 1;
3063     }
3064 
3065     switch (output_format) {
3066     case OFORMAT_HUMAN:
3067         dump_human_image_info_list(list);
3068         break;
3069     case OFORMAT_JSON:
3070         if (chain) {
3071             dump_json_block_graph_info_list(list);
3072         } else {
3073             dump_json_block_graph_info(list->value);
3074         }
3075         break;
3076     }
3077 
3078     qapi_free_BlockGraphInfoList(list);
3079     return 0;
3080 }
3081 
3082 static int dump_map_entry(OutputFormat output_format, MapEntry *e,
3083                           MapEntry *next)
3084 {
3085     switch (output_format) {
3086     case OFORMAT_HUMAN:
3087         if (e->data && !e->has_offset) {
3088             error_report("File contains external, encrypted or compressed clusters.");
3089             return -1;
3090         }
3091         if (e->data && !e->zero) {
3092             printf("%#-16"PRIx64"%#-16"PRIx64"%#-16"PRIx64"%s\n",
3093                    e->start, e->length,
3094                    e->has_offset ? e->offset : 0,
3095                    e->filename ?: "");
3096         }
3097         /* This format ignores the distinction between 0, ZERO and ZERO|DATA.
3098          * Modify the flags here to allow more coalescing.
3099          */
3100         if (next && (!next->data || next->zero)) {
3101             next->data = false;
3102             next->zero = true;
3103         }
3104         break;
3105     case OFORMAT_JSON:
3106         printf("{ \"start\": %"PRId64", \"length\": %"PRId64","
3107                " \"depth\": %"PRId64", \"present\": %s, \"zero\": %s,"
3108                " \"data\": %s", e->start, e->length, e->depth,
3109                e->present ? "true" : "false",
3110                e->zero ? "true" : "false",
3111                e->data ? "true" : "false");
3112         if (e->has_offset) {
3113             printf(", \"offset\": %"PRId64"", e->offset);
3114         }
3115         putchar('}');
3116 
3117         if (next) {
3118             puts(",");
3119         }
3120         break;
3121     }
3122     return 0;
3123 }
3124 
3125 static int get_block_status(BlockDriverState *bs, int64_t offset,
3126                             int64_t bytes, MapEntry *e)
3127 {
3128     int ret;
3129     int depth;
3130     BlockDriverState *file;
3131     bool has_offset;
3132     int64_t map;
3133     char *filename = NULL;
3134 
3135     /* As an optimization, we could cache the current range of unallocated
3136      * clusters in each file of the chain, and avoid querying the same
3137      * range repeatedly.
3138      */
3139 
3140     depth = 0;
3141     for (;;) {
3142         bs = bdrv_skip_filters(bs);
3143         ret = bdrv_block_status(bs, offset, bytes, &bytes, &map, &file);
3144         if (ret < 0) {
3145             return ret;
3146         }
3147         assert(bytes);
3148         if (ret & (BDRV_BLOCK_ZERO|BDRV_BLOCK_DATA)) {
3149             break;
3150         }
3151         bs = bdrv_cow_bs(bs);
3152         if (bs == NULL) {
3153             ret = 0;
3154             break;
3155         }
3156 
3157         depth++;
3158     }
3159 
3160     has_offset = !!(ret & BDRV_BLOCK_OFFSET_VALID);
3161 
3162     if (file && has_offset) {
3163         bdrv_refresh_filename(file);
3164         filename = file->filename;
3165     }
3166 
3167     *e = (MapEntry) {
3168         .start = offset,
3169         .length = bytes,
3170         .data = !!(ret & BDRV_BLOCK_DATA),
3171         .zero = !!(ret & BDRV_BLOCK_ZERO),
3172         .offset = map,
3173         .has_offset = has_offset,
3174         .depth = depth,
3175         .present = !!(ret & BDRV_BLOCK_ALLOCATED),
3176         .filename = filename,
3177     };
3178 
3179     return 0;
3180 }
3181 
3182 static inline bool entry_mergeable(const MapEntry *curr, const MapEntry *next)
3183 {
3184     if (curr->length == 0) {
3185         return false;
3186     }
3187     if (curr->zero != next->zero ||
3188         curr->data != next->data ||
3189         curr->depth != next->depth ||
3190         curr->present != next->present ||
3191         !curr->filename != !next->filename ||
3192         curr->has_offset != next->has_offset) {
3193         return false;
3194     }
3195     if (curr->filename && strcmp(curr->filename, next->filename)) {
3196         return false;
3197     }
3198     if (curr->has_offset && curr->offset + curr->length != next->offset) {
3199         return false;
3200     }
3201     return true;
3202 }
3203 
3204 static int img_map(int argc, char **argv)
3205 {
3206     int c;
3207     OutputFormat output_format = OFORMAT_HUMAN;
3208     BlockBackend *blk;
3209     BlockDriverState *bs;
3210     const char *filename, *fmt, *output;
3211     int64_t length;
3212     MapEntry curr = { .length = 0 }, next;
3213     int ret = 0;
3214     bool image_opts = false;
3215     bool force_share = false;
3216     int64_t start_offset = 0;
3217     int64_t max_length = -1;
3218 
3219     fmt = NULL;
3220     output = NULL;
3221     for (;;) {
3222         int option_index = 0;
3223         static const struct option long_options[] = {
3224             {"help", no_argument, 0, 'h'},
3225             {"format", required_argument, 0, 'f'},
3226             {"output", required_argument, 0, OPTION_OUTPUT},
3227             {"object", required_argument, 0, OPTION_OBJECT},
3228             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3229             {"force-share", no_argument, 0, 'U'},
3230             {"start-offset", required_argument, 0, 's'},
3231             {"max-length", required_argument, 0, 'l'},
3232             {0, 0, 0, 0}
3233         };
3234         c = getopt_long(argc, argv, ":f:s:l:hU",
3235                         long_options, &option_index);
3236         if (c == -1) {
3237             break;
3238         }
3239         switch (c) {
3240         case ':':
3241             missing_argument(argv[optind - 1]);
3242             break;
3243         case '?':
3244             unrecognized_option(argv[optind - 1]);
3245             break;
3246         case 'h':
3247             help();
3248             break;
3249         case 'f':
3250             fmt = optarg;
3251             break;
3252         case 'U':
3253             force_share = true;
3254             break;
3255         case OPTION_OUTPUT:
3256             output = optarg;
3257             break;
3258         case 's':
3259             start_offset = cvtnum("start offset", optarg);
3260             if (start_offset < 0) {
3261                 return 1;
3262             }
3263             break;
3264         case 'l':
3265             max_length = cvtnum("max length", optarg);
3266             if (max_length < 0) {
3267                 return 1;
3268             }
3269             break;
3270         case OPTION_OBJECT:
3271             user_creatable_process_cmdline(optarg);
3272             break;
3273         case OPTION_IMAGE_OPTS:
3274             image_opts = true;
3275             break;
3276         }
3277     }
3278     if (optind != argc - 1) {
3279         error_exit("Expecting one image file name");
3280     }
3281     filename = argv[optind];
3282 
3283     if (output && !strcmp(output, "json")) {
3284         output_format = OFORMAT_JSON;
3285     } else if (output && !strcmp(output, "human")) {
3286         output_format = OFORMAT_HUMAN;
3287     } else if (output) {
3288         error_report("--output must be used with human or json as argument.");
3289         return 1;
3290     }
3291 
3292     blk = img_open(image_opts, filename, fmt, 0, false, false, force_share);
3293     if (!blk) {
3294         return 1;
3295     }
3296     bs = blk_bs(blk);
3297 
3298     if (output_format == OFORMAT_HUMAN) {
3299         printf("%-16s%-16s%-16s%s\n", "Offset", "Length", "Mapped to", "File");
3300     } else if (output_format == OFORMAT_JSON) {
3301         putchar('[');
3302     }
3303 
3304     length = blk_getlength(blk);
3305     if (length < 0) {
3306         error_report("Failed to get size for '%s'", filename);
3307         return 1;
3308     }
3309     if (max_length != -1) {
3310         length = MIN(start_offset + max_length, length);
3311     }
3312 
3313     curr.start = start_offset;
3314     while (curr.start + curr.length < length) {
3315         int64_t offset = curr.start + curr.length;
3316         int64_t n = length - offset;
3317 
3318         ret = get_block_status(bs, offset, n, &next);
3319         if (ret < 0) {
3320             error_report("Could not read file metadata: %s", strerror(-ret));
3321             goto out;
3322         }
3323 
3324         if (entry_mergeable(&curr, &next)) {
3325             curr.length += next.length;
3326             continue;
3327         }
3328 
3329         if (curr.length > 0) {
3330             ret = dump_map_entry(output_format, &curr, &next);
3331             if (ret < 0) {
3332                 goto out;
3333             }
3334         }
3335         curr = next;
3336     }
3337 
3338     ret = dump_map_entry(output_format, &curr, NULL);
3339     if (output_format == OFORMAT_JSON) {
3340         puts("]");
3341     }
3342 
3343 out:
3344     blk_unref(blk);
3345     return ret < 0;
3346 }
3347 
3348 #define SNAPSHOT_LIST   1
3349 #define SNAPSHOT_CREATE 2
3350 #define SNAPSHOT_APPLY  3
3351 #define SNAPSHOT_DELETE 4
3352 
3353 static int img_snapshot(int argc, char **argv)
3354 {
3355     BlockBackend *blk;
3356     BlockDriverState *bs;
3357     QEMUSnapshotInfo sn;
3358     char *filename, *snapshot_name = NULL;
3359     int c, ret = 0, bdrv_oflags;
3360     int action = 0;
3361     bool quiet = false;
3362     Error *err = NULL;
3363     bool image_opts = false;
3364     bool force_share = false;
3365     int64_t rt;
3366 
3367     bdrv_oflags = BDRV_O_RDWR;
3368     /* Parse commandline parameters */
3369     for(;;) {
3370         static const struct option long_options[] = {
3371             {"help", no_argument, 0, 'h'},
3372             {"object", required_argument, 0, OPTION_OBJECT},
3373             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3374             {"force-share", no_argument, 0, 'U'},
3375             {0, 0, 0, 0}
3376         };
3377         c = getopt_long(argc, argv, ":la:c:d:hqU",
3378                         long_options, NULL);
3379         if (c == -1) {
3380             break;
3381         }
3382         switch(c) {
3383         case ':':
3384             missing_argument(argv[optind - 1]);
3385             break;
3386         case '?':
3387             unrecognized_option(argv[optind - 1]);
3388             break;
3389         case 'h':
3390             help();
3391             return 0;
3392         case 'l':
3393             if (action) {
3394                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3395                 return 0;
3396             }
3397             action = SNAPSHOT_LIST;
3398             bdrv_oflags &= ~BDRV_O_RDWR; /* no need for RW */
3399             break;
3400         case 'a':
3401             if (action) {
3402                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3403                 return 0;
3404             }
3405             action = SNAPSHOT_APPLY;
3406             snapshot_name = optarg;
3407             break;
3408         case 'c':
3409             if (action) {
3410                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3411                 return 0;
3412             }
3413             action = SNAPSHOT_CREATE;
3414             snapshot_name = optarg;
3415             break;
3416         case 'd':
3417             if (action) {
3418                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
3419                 return 0;
3420             }
3421             action = SNAPSHOT_DELETE;
3422             snapshot_name = optarg;
3423             break;
3424         case 'q':
3425             quiet = true;
3426             break;
3427         case 'U':
3428             force_share = true;
3429             break;
3430         case OPTION_OBJECT:
3431             user_creatable_process_cmdline(optarg);
3432             break;
3433         case OPTION_IMAGE_OPTS:
3434             image_opts = true;
3435             break;
3436         }
3437     }
3438 
3439     if (optind != argc - 1) {
3440         error_exit("Expecting one image file name");
3441     }
3442     filename = argv[optind++];
3443 
3444     /* Open the image */
3445     blk = img_open(image_opts, filename, NULL, bdrv_oflags, false, quiet,
3446                    force_share);
3447     if (!blk) {
3448         return 1;
3449     }
3450     bs = blk_bs(blk);
3451 
3452     /* Perform the requested action */
3453     switch(action) {
3454     case SNAPSHOT_LIST:
3455         dump_snapshots(bs);
3456         break;
3457 
3458     case SNAPSHOT_CREATE:
3459         memset(&sn, 0, sizeof(sn));
3460         pstrcpy(sn.name, sizeof(sn.name), snapshot_name);
3461 
3462         rt = g_get_real_time();
3463         sn.date_sec = rt / G_USEC_PER_SEC;
3464         sn.date_nsec = (rt % G_USEC_PER_SEC) * 1000;
3465 
3466         ret = bdrv_snapshot_create(bs, &sn);
3467         if (ret) {
3468             error_report("Could not create snapshot '%s': %d (%s)",
3469                 snapshot_name, ret, strerror(-ret));
3470         }
3471         break;
3472 
3473     case SNAPSHOT_APPLY:
3474         ret = bdrv_snapshot_goto(bs, snapshot_name, &err);
3475         if (ret) {
3476             error_reportf_err(err, "Could not apply snapshot '%s': ",
3477                               snapshot_name);
3478         }
3479         break;
3480 
3481     case SNAPSHOT_DELETE:
3482         ret = bdrv_snapshot_find(bs, &sn, snapshot_name);
3483         if (ret < 0) {
3484             error_report("Could not delete snapshot '%s': snapshot not "
3485                          "found", snapshot_name);
3486             ret = 1;
3487         } else {
3488             ret = bdrv_snapshot_delete(bs, sn.id_str, sn.name, &err);
3489             if (ret < 0) {
3490                 error_reportf_err(err, "Could not delete snapshot '%s': ",
3491                                   snapshot_name);
3492                 ret = 1;
3493             }
3494         }
3495         break;
3496     }
3497 
3498     /* Cleanup */
3499     blk_unref(blk);
3500     if (ret) {
3501         return 1;
3502     }
3503     return 0;
3504 }
3505 
3506 static int img_rebase(int argc, char **argv)
3507 {
3508     BlockBackend *blk = NULL, *blk_old_backing = NULL, *blk_new_backing = NULL;
3509     uint8_t *buf_old = NULL;
3510     uint8_t *buf_new = NULL;
3511     BlockDriverState *bs = NULL, *prefix_chain_bs = NULL;
3512     BlockDriverState *unfiltered_bs;
3513     char *filename;
3514     const char *fmt, *cache, *src_cache, *out_basefmt, *out_baseimg;
3515     int c, flags, src_flags, ret;
3516     bool writethrough, src_writethrough;
3517     int unsafe = 0;
3518     bool force_share = false;
3519     int progress = 0;
3520     bool quiet = false;
3521     Error *local_err = NULL;
3522     bool image_opts = false;
3523 
3524     /* Parse commandline parameters */
3525     fmt = NULL;
3526     cache = BDRV_DEFAULT_CACHE;
3527     src_cache = BDRV_DEFAULT_CACHE;
3528     out_baseimg = NULL;
3529     out_basefmt = NULL;
3530     for(;;) {
3531         static const struct option long_options[] = {
3532             {"help", no_argument, 0, 'h'},
3533             {"object", required_argument, 0, OPTION_OBJECT},
3534             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3535             {"force-share", no_argument, 0, 'U'},
3536             {0, 0, 0, 0}
3537         };
3538         c = getopt_long(argc, argv, ":hf:F:b:upt:T:qU",
3539                         long_options, NULL);
3540         if (c == -1) {
3541             break;
3542         }
3543         switch(c) {
3544         case ':':
3545             missing_argument(argv[optind - 1]);
3546             break;
3547         case '?':
3548             unrecognized_option(argv[optind - 1]);
3549             break;
3550         case 'h':
3551             help();
3552             return 0;
3553         case 'f':
3554             fmt = optarg;
3555             break;
3556         case 'F':
3557             out_basefmt = optarg;
3558             break;
3559         case 'b':
3560             out_baseimg = optarg;
3561             break;
3562         case 'u':
3563             unsafe = 1;
3564             break;
3565         case 'p':
3566             progress = 1;
3567             break;
3568         case 't':
3569             cache = optarg;
3570             break;
3571         case 'T':
3572             src_cache = optarg;
3573             break;
3574         case 'q':
3575             quiet = true;
3576             break;
3577         case OPTION_OBJECT:
3578             user_creatable_process_cmdline(optarg);
3579             break;
3580         case OPTION_IMAGE_OPTS:
3581             image_opts = true;
3582             break;
3583         case 'U':
3584             force_share = true;
3585             break;
3586         }
3587     }
3588 
3589     if (quiet) {
3590         progress = 0;
3591     }
3592 
3593     if (optind != argc - 1) {
3594         error_exit("Expecting one image file name");
3595     }
3596     if (!unsafe && !out_baseimg) {
3597         error_exit("Must specify backing file (-b) or use unsafe mode (-u)");
3598     }
3599     filename = argv[optind++];
3600 
3601     qemu_progress_init(progress, 2.0);
3602     qemu_progress_print(0, 100);
3603 
3604     flags = BDRV_O_RDWR | (unsafe ? BDRV_O_NO_BACKING : 0);
3605     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
3606     if (ret < 0) {
3607         error_report("Invalid cache option: %s", cache);
3608         goto out;
3609     }
3610 
3611     src_flags = 0;
3612     ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
3613     if (ret < 0) {
3614         error_report("Invalid source cache option: %s", src_cache);
3615         goto out;
3616     }
3617 
3618     /* The source files are opened read-only, don't care about WCE */
3619     assert((src_flags & BDRV_O_RDWR) == 0);
3620     (void) src_writethrough;
3621 
3622     /*
3623      * Open the images.
3624      *
3625      * Ignore the old backing file for unsafe rebase in case we want to correct
3626      * the reference to a renamed or moved backing file.
3627      */
3628     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
3629                    false);
3630     if (!blk) {
3631         ret = -1;
3632         goto out;
3633     }
3634     bs = blk_bs(blk);
3635 
3636     unfiltered_bs = bdrv_skip_filters(bs);
3637 
3638     if (out_basefmt != NULL) {
3639         if (bdrv_find_format(out_basefmt) == NULL) {
3640             error_report("Invalid format name: '%s'", out_basefmt);
3641             ret = -1;
3642             goto out;
3643         }
3644     }
3645 
3646     /* For safe rebasing we need to compare old and new backing file */
3647     if (!unsafe) {
3648         QDict *options = NULL;
3649         BlockDriverState *base_bs = bdrv_cow_bs(unfiltered_bs);
3650 
3651         if (base_bs) {
3652             blk_old_backing = blk_new(qemu_get_aio_context(),
3653                                       BLK_PERM_CONSISTENT_READ,
3654                                       BLK_PERM_ALL);
3655             ret = blk_insert_bs(blk_old_backing, base_bs,
3656                                 &local_err);
3657             if (ret < 0) {
3658                 error_reportf_err(local_err,
3659                                   "Could not reuse old backing file '%s': ",
3660                                   base_bs->filename);
3661                 goto out;
3662             }
3663         } else {
3664             blk_old_backing = NULL;
3665         }
3666 
3667         if (out_baseimg[0]) {
3668             const char *overlay_filename;
3669             char *out_real_path;
3670 
3671             options = qdict_new();
3672             if (out_basefmt) {
3673                 qdict_put_str(options, "driver", out_basefmt);
3674             }
3675             if (force_share) {
3676                 qdict_put_bool(options, BDRV_OPT_FORCE_SHARE, true);
3677             }
3678 
3679             bdrv_refresh_filename(bs);
3680             overlay_filename = bs->exact_filename[0] ? bs->exact_filename
3681                                                      : bs->filename;
3682             out_real_path =
3683                 bdrv_get_full_backing_filename_from_filename(overlay_filename,
3684                                                              out_baseimg,
3685                                                              &local_err);
3686             if (local_err) {
3687                 qobject_unref(options);
3688                 error_reportf_err(local_err,
3689                                   "Could not resolve backing filename: ");
3690                 ret = -1;
3691                 goto out;
3692             }
3693 
3694             /*
3695              * Find out whether we rebase an image on top of a previous image
3696              * in its chain.
3697              */
3698             prefix_chain_bs = bdrv_find_backing_image(bs, out_real_path);
3699             if (prefix_chain_bs) {
3700                 qobject_unref(options);
3701                 g_free(out_real_path);
3702 
3703                 blk_new_backing = blk_new(qemu_get_aio_context(),
3704                                           BLK_PERM_CONSISTENT_READ,
3705                                           BLK_PERM_ALL);
3706                 ret = blk_insert_bs(blk_new_backing, prefix_chain_bs,
3707                                     &local_err);
3708                 if (ret < 0) {
3709                     error_reportf_err(local_err,
3710                                       "Could not reuse backing file '%s': ",
3711                                       out_baseimg);
3712                     goto out;
3713                 }
3714             } else {
3715                 blk_new_backing = blk_new_open(out_real_path, NULL,
3716                                                options, src_flags, &local_err);
3717                 g_free(out_real_path);
3718                 if (!blk_new_backing) {
3719                     error_reportf_err(local_err,
3720                                       "Could not open new backing file '%s': ",
3721                                       out_baseimg);
3722                     ret = -1;
3723                     goto out;
3724                 }
3725             }
3726         }
3727     }
3728 
3729     /*
3730      * Check each unallocated cluster in the COW file. If it is unallocated,
3731      * accesses go to the backing file. We must therefore compare this cluster
3732      * in the old and new backing file, and if they differ we need to copy it
3733      * from the old backing file into the COW file.
3734      *
3735      * If qemu-img crashes during this step, no harm is done. The content of
3736      * the image is the same as the original one at any time.
3737      */
3738     if (!unsafe) {
3739         int64_t size;
3740         int64_t old_backing_size = 0;
3741         int64_t new_backing_size = 0;
3742         uint64_t offset;
3743         int64_t n;
3744         float local_progress = 0;
3745 
3746         buf_old = blk_blockalign(blk, IO_BUF_SIZE);
3747         buf_new = blk_blockalign(blk, IO_BUF_SIZE);
3748 
3749         size = blk_getlength(blk);
3750         if (size < 0) {
3751             error_report("Could not get size of '%s': %s",
3752                          filename, strerror(-size));
3753             ret = -1;
3754             goto out;
3755         }
3756         if (blk_old_backing) {
3757             old_backing_size = blk_getlength(blk_old_backing);
3758             if (old_backing_size < 0) {
3759                 char backing_name[PATH_MAX];
3760 
3761                 bdrv_get_backing_filename(bs, backing_name,
3762                                           sizeof(backing_name));
3763                 error_report("Could not get size of '%s': %s",
3764                              backing_name, strerror(-old_backing_size));
3765                 ret = -1;
3766                 goto out;
3767             }
3768         }
3769         if (blk_new_backing) {
3770             new_backing_size = blk_getlength(blk_new_backing);
3771             if (new_backing_size < 0) {
3772                 error_report("Could not get size of '%s': %s",
3773                              out_baseimg, strerror(-new_backing_size));
3774                 ret = -1;
3775                 goto out;
3776             }
3777         }
3778 
3779         if (size != 0) {
3780             local_progress = (float)100 / (size / MIN(size, IO_BUF_SIZE));
3781         }
3782 
3783         for (offset = 0; offset < size; offset += n) {
3784             bool buf_old_is_zero = false;
3785 
3786             /* How many bytes can we handle with the next read? */
3787             n = MIN(IO_BUF_SIZE, size - offset);
3788 
3789             /* If the cluster is allocated, we don't need to take action */
3790             ret = bdrv_is_allocated(unfiltered_bs, offset, n, &n);
3791             if (ret < 0) {
3792                 error_report("error while reading image metadata: %s",
3793                              strerror(-ret));
3794                 goto out;
3795             }
3796             if (ret) {
3797                 continue;
3798             }
3799 
3800             if (prefix_chain_bs) {
3801                 /*
3802                  * If cluster wasn't changed since prefix_chain, we don't need
3803                  * to take action
3804                  */
3805                 ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
3806                                               prefix_chain_bs, false,
3807                                               offset, n, &n);
3808                 if (ret < 0) {
3809                     error_report("error while reading image metadata: %s",
3810                                  strerror(-ret));
3811                     goto out;
3812                 }
3813                 if (!ret) {
3814                     continue;
3815                 }
3816             }
3817 
3818             /*
3819              * Read old and new backing file and take into consideration that
3820              * backing files may be smaller than the COW image.
3821              */
3822             if (offset >= old_backing_size) {
3823                 memset(buf_old, 0, n);
3824                 buf_old_is_zero = true;
3825             } else {
3826                 if (offset + n > old_backing_size) {
3827                     n = old_backing_size - offset;
3828                 }
3829 
3830                 ret = blk_pread(blk_old_backing, offset, n, buf_old, 0);
3831                 if (ret < 0) {
3832                     error_report("error while reading from old backing file");
3833                     goto out;
3834                 }
3835             }
3836 
3837             if (offset >= new_backing_size || !blk_new_backing) {
3838                 memset(buf_new, 0, n);
3839             } else {
3840                 if (offset + n > new_backing_size) {
3841                     n = new_backing_size - offset;
3842                 }
3843 
3844                 ret = blk_pread(blk_new_backing, offset, n, buf_new, 0);
3845                 if (ret < 0) {
3846                     error_report("error while reading from new backing file");
3847                     goto out;
3848                 }
3849             }
3850 
3851             /* If they differ, we need to write to the COW file */
3852             uint64_t written = 0;
3853 
3854             while (written < n) {
3855                 int64_t pnum;
3856 
3857                 if (compare_buffers(buf_old + written, buf_new + written,
3858                                     n - written, &pnum))
3859                 {
3860                     if (buf_old_is_zero) {
3861                         ret = blk_pwrite_zeroes(blk, offset + written, pnum, 0);
3862                     } else {
3863                         ret = blk_pwrite(blk, offset + written, pnum,
3864                                          buf_old + written, 0);
3865                     }
3866                     if (ret < 0) {
3867                         error_report("Error while writing to COW image: %s",
3868                             strerror(-ret));
3869                         goto out;
3870                     }
3871                 }
3872 
3873                 written += pnum;
3874             }
3875             qemu_progress_print(local_progress, 100);
3876         }
3877     }
3878 
3879     /*
3880      * Change the backing file. All clusters that are different from the old
3881      * backing file are overwritten in the COW file now, so the visible content
3882      * doesn't change when we switch the backing file.
3883      */
3884     if (out_baseimg && *out_baseimg) {
3885         ret = bdrv_change_backing_file(unfiltered_bs, out_baseimg, out_basefmt,
3886                                        true);
3887     } else {
3888         ret = bdrv_change_backing_file(unfiltered_bs, NULL, NULL, false);
3889     }
3890 
3891     if (ret == -ENOSPC) {
3892         error_report("Could not change the backing file to '%s': No "
3893                      "space left in the file header", out_baseimg);
3894     } else if (ret == -EINVAL && out_baseimg && !out_basefmt) {
3895         error_report("Could not change the backing file to '%s': backing "
3896                      "format must be specified", out_baseimg);
3897     } else if (ret < 0) {
3898         error_report("Could not change the backing file to '%s': %s",
3899             out_baseimg, strerror(-ret));
3900     }
3901 
3902     qemu_progress_print(100, 0);
3903     /*
3904      * TODO At this point it is possible to check if any clusters that are
3905      * allocated in the COW file are the same in the backing file. If so, they
3906      * could be dropped from the COW file. Don't do this before switching the
3907      * backing file, in case of a crash this would lead to corruption.
3908      */
3909 out:
3910     qemu_progress_end();
3911     /* Cleanup */
3912     if (!unsafe) {
3913         blk_unref(blk_old_backing);
3914         blk_unref(blk_new_backing);
3915     }
3916     qemu_vfree(buf_old);
3917     qemu_vfree(buf_new);
3918 
3919     blk_unref(blk);
3920     if (ret) {
3921         return 1;
3922     }
3923     return 0;
3924 }
3925 
3926 static int img_resize(int argc, char **argv)
3927 {
3928     Error *err = NULL;
3929     int c, ret, relative;
3930     const char *filename, *fmt, *size;
3931     int64_t n, total_size, current_size;
3932     bool quiet = false;
3933     BlockBackend *blk = NULL;
3934     PreallocMode prealloc = PREALLOC_MODE_OFF;
3935     QemuOpts *param;
3936 
3937     static QemuOptsList resize_options = {
3938         .name = "resize_options",
3939         .head = QTAILQ_HEAD_INITIALIZER(resize_options.head),
3940         .desc = {
3941             {
3942                 .name = BLOCK_OPT_SIZE,
3943                 .type = QEMU_OPT_SIZE,
3944                 .help = "Virtual disk size"
3945             }, {
3946                 /* end of list */
3947             }
3948         },
3949     };
3950     bool image_opts = false;
3951     bool shrink = false;
3952 
3953     /* Remove size from argv manually so that negative numbers are not treated
3954      * as options by getopt. */
3955     if (argc < 3) {
3956         error_exit("Not enough arguments");
3957         return 1;
3958     }
3959 
3960     size = argv[--argc];
3961 
3962     /* Parse getopt arguments */
3963     fmt = NULL;
3964     for(;;) {
3965         static const struct option long_options[] = {
3966             {"help", no_argument, 0, 'h'},
3967             {"object", required_argument, 0, OPTION_OBJECT},
3968             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3969             {"preallocation", required_argument, 0, OPTION_PREALLOCATION},
3970             {"shrink", no_argument, 0, OPTION_SHRINK},
3971             {0, 0, 0, 0}
3972         };
3973         c = getopt_long(argc, argv, ":f:hq",
3974                         long_options, NULL);
3975         if (c == -1) {
3976             break;
3977         }
3978         switch(c) {
3979         case ':':
3980             missing_argument(argv[optind - 1]);
3981             break;
3982         case '?':
3983             unrecognized_option(argv[optind - 1]);
3984             break;
3985         case 'h':
3986             help();
3987             break;
3988         case 'f':
3989             fmt = optarg;
3990             break;
3991         case 'q':
3992             quiet = true;
3993             break;
3994         case OPTION_OBJECT:
3995             user_creatable_process_cmdline(optarg);
3996             break;
3997         case OPTION_IMAGE_OPTS:
3998             image_opts = true;
3999             break;
4000         case OPTION_PREALLOCATION:
4001             prealloc = qapi_enum_parse(&PreallocMode_lookup, optarg,
4002                                        PREALLOC_MODE__MAX, NULL);
4003             if (prealloc == PREALLOC_MODE__MAX) {
4004                 error_report("Invalid preallocation mode '%s'", optarg);
4005                 return 1;
4006             }
4007             break;
4008         case OPTION_SHRINK:
4009             shrink = true;
4010             break;
4011         }
4012     }
4013     if (optind != argc - 1) {
4014         error_exit("Expecting image file name and size");
4015     }
4016     filename = argv[optind++];
4017 
4018     /* Choose grow, shrink, or absolute resize mode */
4019     switch (size[0]) {
4020     case '+':
4021         relative = 1;
4022         size++;
4023         break;
4024     case '-':
4025         relative = -1;
4026         size++;
4027         break;
4028     default:
4029         relative = 0;
4030         break;
4031     }
4032 
4033     /* Parse size */
4034     param = qemu_opts_create(&resize_options, NULL, 0, &error_abort);
4035     if (!qemu_opt_set(param, BLOCK_OPT_SIZE, size, &err)) {
4036         error_report_err(err);
4037         ret = -1;
4038         qemu_opts_del(param);
4039         goto out;
4040     }
4041     n = qemu_opt_get_size(param, BLOCK_OPT_SIZE, 0);
4042     qemu_opts_del(param);
4043 
4044     blk = img_open(image_opts, filename, fmt,
4045                    BDRV_O_RDWR | BDRV_O_RESIZE, false, quiet,
4046                    false);
4047     if (!blk) {
4048         ret = -1;
4049         goto out;
4050     }
4051 
4052     current_size = blk_getlength(blk);
4053     if (current_size < 0) {
4054         error_report("Failed to inquire current image length: %s",
4055                      strerror(-current_size));
4056         ret = -1;
4057         goto out;
4058     }
4059 
4060     if (relative) {
4061         total_size = current_size + n * relative;
4062     } else {
4063         total_size = n;
4064     }
4065     if (total_size <= 0) {
4066         error_report("New image size must be positive");
4067         ret = -1;
4068         goto out;
4069     }
4070 
4071     if (total_size <= current_size && prealloc != PREALLOC_MODE_OFF) {
4072         error_report("Preallocation can only be used for growing images");
4073         ret = -1;
4074         goto out;
4075     }
4076 
4077     if (total_size < current_size && !shrink) {
4078         error_report("Use the --shrink option to perform a shrink operation.");
4079         warn_report("Shrinking an image will delete all data beyond the "
4080                     "shrunken image's end. Before performing such an "
4081                     "operation, make sure there is no important data there.");
4082         ret = -1;
4083         goto out;
4084     }
4085 
4086     /*
4087      * The user expects the image to have the desired size after
4088      * resizing, so pass @exact=true.  It is of no use to report
4089      * success when the image has not actually been resized.
4090      */
4091     ret = blk_truncate(blk, total_size, true, prealloc, 0, &err);
4092     if (!ret) {
4093         qprintf(quiet, "Image resized.\n");
4094     } else {
4095         error_report_err(err);
4096     }
4097 out:
4098     blk_unref(blk);
4099     if (ret) {
4100         return 1;
4101     }
4102     return 0;
4103 }
4104 
4105 static void amend_status_cb(BlockDriverState *bs,
4106                             int64_t offset, int64_t total_work_size,
4107                             void *opaque)
4108 {
4109     qemu_progress_print(100.f * offset / total_work_size, 0);
4110 }
4111 
4112 static int print_amend_option_help(const char *format)
4113 {
4114     BlockDriver *drv;
4115 
4116     /* Find driver and parse its options */
4117     drv = bdrv_find_format(format);
4118     if (!drv) {
4119         error_report("Unknown file format '%s'", format);
4120         return 1;
4121     }
4122 
4123     if (!drv->bdrv_amend_options) {
4124         error_report("Format driver '%s' does not support option amendment",
4125                      format);
4126         return 1;
4127     }
4128 
4129     /* Every driver supporting amendment must have amend_opts */
4130     assert(drv->amend_opts);
4131 
4132     printf("Amend options for '%s':\n", format);
4133     qemu_opts_print_help(drv->amend_opts, false);
4134     return 0;
4135 }
4136 
4137 static int img_amend(int argc, char **argv)
4138 {
4139     Error *err = NULL;
4140     int c, ret = 0;
4141     char *options = NULL;
4142     QemuOptsList *amend_opts = NULL;
4143     QemuOpts *opts = NULL;
4144     const char *fmt = NULL, *filename, *cache;
4145     int flags;
4146     bool writethrough;
4147     bool quiet = false, progress = false;
4148     BlockBackend *blk = NULL;
4149     BlockDriverState *bs = NULL;
4150     bool image_opts = false;
4151     bool force = false;
4152 
4153     cache = BDRV_DEFAULT_CACHE;
4154     for (;;) {
4155         static const struct option long_options[] = {
4156             {"help", no_argument, 0, 'h'},
4157             {"object", required_argument, 0, OPTION_OBJECT},
4158             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4159             {"force", no_argument, 0, OPTION_FORCE},
4160             {0, 0, 0, 0}
4161         };
4162         c = getopt_long(argc, argv, ":ho:f:t:pq",
4163                         long_options, NULL);
4164         if (c == -1) {
4165             break;
4166         }
4167 
4168         switch (c) {
4169         case ':':
4170             missing_argument(argv[optind - 1]);
4171             break;
4172         case '?':
4173             unrecognized_option(argv[optind - 1]);
4174             break;
4175         case 'h':
4176             help();
4177             break;
4178         case 'o':
4179             if (accumulate_options(&options, optarg) < 0) {
4180                 ret = -1;
4181                 goto out_no_progress;
4182             }
4183             break;
4184         case 'f':
4185             fmt = optarg;
4186             break;
4187         case 't':
4188             cache = optarg;
4189             break;
4190         case 'p':
4191             progress = true;
4192             break;
4193         case 'q':
4194             quiet = true;
4195             break;
4196         case OPTION_OBJECT:
4197             user_creatable_process_cmdline(optarg);
4198             break;
4199         case OPTION_IMAGE_OPTS:
4200             image_opts = true;
4201             break;
4202         case OPTION_FORCE:
4203             force = true;
4204             break;
4205         }
4206     }
4207 
4208     if (!options) {
4209         error_exit("Must specify options (-o)");
4210     }
4211 
4212     if (quiet) {
4213         progress = false;
4214     }
4215     qemu_progress_init(progress, 1.0);
4216 
4217     filename = (optind == argc - 1) ? argv[argc - 1] : NULL;
4218     if (fmt && has_help_option(options)) {
4219         /* If a format is explicitly specified (and possibly no filename is
4220          * given), print option help here */
4221         ret = print_amend_option_help(fmt);
4222         goto out;
4223     }
4224 
4225     if (optind != argc - 1) {
4226         error_report("Expecting one image file name");
4227         ret = -1;
4228         goto out;
4229     }
4230 
4231     flags = BDRV_O_RDWR;
4232     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
4233     if (ret < 0) {
4234         error_report("Invalid cache option: %s", cache);
4235         goto out;
4236     }
4237 
4238     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
4239                    false);
4240     if (!blk) {
4241         ret = -1;
4242         goto out;
4243     }
4244     bs = blk_bs(blk);
4245 
4246     fmt = bs->drv->format_name;
4247 
4248     if (has_help_option(options)) {
4249         /* If the format was auto-detected, print option help here */
4250         ret = print_amend_option_help(fmt);
4251         goto out;
4252     }
4253 
4254     if (!bs->drv->bdrv_amend_options) {
4255         error_report("Format driver '%s' does not support option amendment",
4256                      fmt);
4257         ret = -1;
4258         goto out;
4259     }
4260 
4261     /* Every driver supporting amendment must have amend_opts */
4262     assert(bs->drv->amend_opts);
4263 
4264     amend_opts = qemu_opts_append(amend_opts, bs->drv->amend_opts);
4265     opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
4266     if (!qemu_opts_do_parse(opts, options, NULL, &err)) {
4267         /* Try to parse options using the create options */
4268         amend_opts = qemu_opts_append(amend_opts, bs->drv->create_opts);
4269         qemu_opts_del(opts);
4270         opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
4271         if (qemu_opts_do_parse(opts, options, NULL, NULL)) {
4272             error_append_hint(&err,
4273                               "This option is only supported for image creation\n");
4274         }
4275 
4276         error_report_err(err);
4277         ret = -1;
4278         goto out;
4279     }
4280 
4281     /* In case the driver does not call amend_status_cb() */
4282     qemu_progress_print(0.f, 0);
4283     ret = bdrv_amend_options(bs, opts, &amend_status_cb, NULL, force, &err);
4284     qemu_progress_print(100.f, 0);
4285     if (ret < 0) {
4286         error_report_err(err);
4287         goto out;
4288     }
4289 
4290 out:
4291     qemu_progress_end();
4292 
4293 out_no_progress:
4294     blk_unref(blk);
4295     qemu_opts_del(opts);
4296     qemu_opts_free(amend_opts);
4297     g_free(options);
4298 
4299     if (ret) {
4300         return 1;
4301     }
4302     return 0;
4303 }
4304 
4305 typedef struct BenchData {
4306     BlockBackend *blk;
4307     uint64_t image_size;
4308     bool write;
4309     int bufsize;
4310     int step;
4311     int nrreq;
4312     int n;
4313     int flush_interval;
4314     bool drain_on_flush;
4315     uint8_t *buf;
4316     QEMUIOVector *qiov;
4317 
4318     int in_flight;
4319     bool in_flush;
4320     uint64_t offset;
4321 } BenchData;
4322 
4323 static void bench_undrained_flush_cb(void *opaque, int ret)
4324 {
4325     if (ret < 0) {
4326         error_report("Failed flush request: %s", strerror(-ret));
4327         exit(EXIT_FAILURE);
4328     }
4329 }
4330 
4331 static void bench_cb(void *opaque, int ret)
4332 {
4333     BenchData *b = opaque;
4334     BlockAIOCB *acb;
4335 
4336     if (ret < 0) {
4337         error_report("Failed request: %s", strerror(-ret));
4338         exit(EXIT_FAILURE);
4339     }
4340 
4341     if (b->in_flush) {
4342         /* Just finished a flush with drained queue: Start next requests */
4343         assert(b->in_flight == 0);
4344         b->in_flush = false;
4345     } else if (b->in_flight > 0) {
4346         int remaining = b->n - b->in_flight;
4347 
4348         b->n--;
4349         b->in_flight--;
4350 
4351         /* Time for flush? Drain queue if requested, then flush */
4352         if (b->flush_interval && remaining % b->flush_interval == 0) {
4353             if (!b->in_flight || !b->drain_on_flush) {
4354                 BlockCompletionFunc *cb;
4355 
4356                 if (b->drain_on_flush) {
4357                     b->in_flush = true;
4358                     cb = bench_cb;
4359                 } else {
4360                     cb = bench_undrained_flush_cb;
4361                 }
4362 
4363                 acb = blk_aio_flush(b->blk, cb, b);
4364                 if (!acb) {
4365                     error_report("Failed to issue flush request");
4366                     exit(EXIT_FAILURE);
4367                 }
4368             }
4369             if (b->drain_on_flush) {
4370                 return;
4371             }
4372         }
4373     }
4374 
4375     while (b->n > b->in_flight && b->in_flight < b->nrreq) {
4376         int64_t offset = b->offset;
4377         /* blk_aio_* might look for completed I/Os and kick bench_cb
4378          * again, so make sure this operation is counted by in_flight
4379          * and b->offset is ready for the next submission.
4380          */
4381         b->in_flight++;
4382         b->offset += b->step;
4383         b->offset %= b->image_size;
4384         if (b->write) {
4385             acb = blk_aio_pwritev(b->blk, offset, b->qiov, 0, bench_cb, b);
4386         } else {
4387             acb = blk_aio_preadv(b->blk, offset, b->qiov, 0, bench_cb, b);
4388         }
4389         if (!acb) {
4390             error_report("Failed to issue request");
4391             exit(EXIT_FAILURE);
4392         }
4393     }
4394 }
4395 
4396 static int img_bench(int argc, char **argv)
4397 {
4398     int c, ret = 0;
4399     const char *fmt = NULL, *filename;
4400     bool quiet = false;
4401     bool image_opts = false;
4402     bool is_write = false;
4403     int count = 75000;
4404     int depth = 64;
4405     int64_t offset = 0;
4406     size_t bufsize = 4096;
4407     int pattern = 0;
4408     size_t step = 0;
4409     int flush_interval = 0;
4410     bool drain_on_flush = true;
4411     int64_t image_size;
4412     BlockBackend *blk = NULL;
4413     BenchData data = {};
4414     int flags = 0;
4415     bool writethrough = false;
4416     struct timeval t1, t2;
4417     int i;
4418     bool force_share = false;
4419     size_t buf_size = 0;
4420 
4421     for (;;) {
4422         static const struct option long_options[] = {
4423             {"help", no_argument, 0, 'h'},
4424             {"flush-interval", required_argument, 0, OPTION_FLUSH_INTERVAL},
4425             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4426             {"pattern", required_argument, 0, OPTION_PATTERN},
4427             {"no-drain", no_argument, 0, OPTION_NO_DRAIN},
4428             {"force-share", no_argument, 0, 'U'},
4429             {0, 0, 0, 0}
4430         };
4431         c = getopt_long(argc, argv, ":hc:d:f:ni:o:qs:S:t:wU", long_options,
4432                         NULL);
4433         if (c == -1) {
4434             break;
4435         }
4436 
4437         switch (c) {
4438         case ':':
4439             missing_argument(argv[optind - 1]);
4440             break;
4441         case '?':
4442             unrecognized_option(argv[optind - 1]);
4443             break;
4444         case 'h':
4445             help();
4446             break;
4447         case 'c':
4448         {
4449             unsigned long res;
4450 
4451             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
4452                 error_report("Invalid request count specified");
4453                 return 1;
4454             }
4455             count = res;
4456             break;
4457         }
4458         case 'd':
4459         {
4460             unsigned long res;
4461 
4462             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
4463                 error_report("Invalid queue depth specified");
4464                 return 1;
4465             }
4466             depth = res;
4467             break;
4468         }
4469         case 'f':
4470             fmt = optarg;
4471             break;
4472         case 'n':
4473             flags |= BDRV_O_NATIVE_AIO;
4474             break;
4475         case 'i':
4476             ret = bdrv_parse_aio(optarg, &flags);
4477             if (ret < 0) {
4478                 error_report("Invalid aio option: %s", optarg);
4479                 ret = -1;
4480                 goto out;
4481             }
4482             break;
4483         case 'o':
4484         {
4485             offset = cvtnum("offset", optarg);
4486             if (offset < 0) {
4487                 return 1;
4488             }
4489             break;
4490         }
4491             break;
4492         case 'q':
4493             quiet = true;
4494             break;
4495         case 's':
4496         {
4497             int64_t sval;
4498 
4499             sval = cvtnum_full("buffer size", optarg, 0, INT_MAX);
4500             if (sval < 0) {
4501                 return 1;
4502             }
4503 
4504             bufsize = sval;
4505             break;
4506         }
4507         case 'S':
4508         {
4509             int64_t sval;
4510 
4511             sval = cvtnum_full("step_size", optarg, 0, INT_MAX);
4512             if (sval < 0) {
4513                 return 1;
4514             }
4515 
4516             step = sval;
4517             break;
4518         }
4519         case 't':
4520             ret = bdrv_parse_cache_mode(optarg, &flags, &writethrough);
4521             if (ret < 0) {
4522                 error_report("Invalid cache mode");
4523                 ret = -1;
4524                 goto out;
4525             }
4526             break;
4527         case 'w':
4528             flags |= BDRV_O_RDWR;
4529             is_write = true;
4530             break;
4531         case 'U':
4532             force_share = true;
4533             break;
4534         case OPTION_PATTERN:
4535         {
4536             unsigned long res;
4537 
4538             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > 0xff) {
4539                 error_report("Invalid pattern byte specified");
4540                 return 1;
4541             }
4542             pattern = res;
4543             break;
4544         }
4545         case OPTION_FLUSH_INTERVAL:
4546         {
4547             unsigned long res;
4548 
4549             if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
4550                 error_report("Invalid flush interval specified");
4551                 return 1;
4552             }
4553             flush_interval = res;
4554             break;
4555         }
4556         case OPTION_NO_DRAIN:
4557             drain_on_flush = false;
4558             break;
4559         case OPTION_IMAGE_OPTS:
4560             image_opts = true;
4561             break;
4562         }
4563     }
4564 
4565     if (optind != argc - 1) {
4566         error_exit("Expecting one image file name");
4567     }
4568     filename = argv[argc - 1];
4569 
4570     if (!is_write && flush_interval) {
4571         error_report("--flush-interval is only available in write tests");
4572         ret = -1;
4573         goto out;
4574     }
4575     if (flush_interval && flush_interval < depth) {
4576         error_report("Flush interval can't be smaller than depth");
4577         ret = -1;
4578         goto out;
4579     }
4580 
4581     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
4582                    force_share);
4583     if (!blk) {
4584         ret = -1;
4585         goto out;
4586     }
4587 
4588     image_size = blk_getlength(blk);
4589     if (image_size < 0) {
4590         ret = image_size;
4591         goto out;
4592     }
4593 
4594     data = (BenchData) {
4595         .blk            = blk,
4596         .image_size     = image_size,
4597         .bufsize        = bufsize,
4598         .step           = step ?: bufsize,
4599         .nrreq          = depth,
4600         .n              = count,
4601         .offset         = offset,
4602         .write          = is_write,
4603         .flush_interval = flush_interval,
4604         .drain_on_flush = drain_on_flush,
4605     };
4606     printf("Sending %d %s requests, %d bytes each, %d in parallel "
4607            "(starting at offset %" PRId64 ", step size %d)\n",
4608            data.n, data.write ? "write" : "read", data.bufsize, data.nrreq,
4609            data.offset, data.step);
4610     if (flush_interval) {
4611         printf("Sending flush every %d requests\n", flush_interval);
4612     }
4613 
4614     buf_size = data.nrreq * data.bufsize;
4615     data.buf = blk_blockalign(blk, buf_size);
4616     memset(data.buf, pattern, data.nrreq * data.bufsize);
4617 
4618     blk_register_buf(blk, data.buf, buf_size, &error_fatal);
4619 
4620     data.qiov = g_new(QEMUIOVector, data.nrreq);
4621     for (i = 0; i < data.nrreq; i++) {
4622         qemu_iovec_init(&data.qiov[i], 1);
4623         qemu_iovec_add(&data.qiov[i],
4624                        data.buf + i * data.bufsize, data.bufsize);
4625     }
4626 
4627     gettimeofday(&t1, NULL);
4628     bench_cb(&data, 0);
4629 
4630     while (data.n > 0) {
4631         main_loop_wait(false);
4632     }
4633     gettimeofday(&t2, NULL);
4634 
4635     printf("Run completed in %3.3f seconds.\n",
4636            (t2.tv_sec - t1.tv_sec)
4637            + ((double)(t2.tv_usec - t1.tv_usec) / 1000000));
4638 
4639 out:
4640     if (data.buf) {
4641         blk_unregister_buf(blk, data.buf, buf_size);
4642     }
4643     qemu_vfree(data.buf);
4644     blk_unref(blk);
4645 
4646     if (ret) {
4647         return 1;
4648     }
4649     return 0;
4650 }
4651 
4652 enum ImgBitmapAct {
4653     BITMAP_ADD,
4654     BITMAP_REMOVE,
4655     BITMAP_CLEAR,
4656     BITMAP_ENABLE,
4657     BITMAP_DISABLE,
4658     BITMAP_MERGE,
4659 };
4660 typedef struct ImgBitmapAction {
4661     enum ImgBitmapAct act;
4662     const char *src; /* only used for merge */
4663     QSIMPLEQ_ENTRY(ImgBitmapAction) next;
4664 } ImgBitmapAction;
4665 
4666 static int img_bitmap(int argc, char **argv)
4667 {
4668     Error *err = NULL;
4669     int c, ret = 1;
4670     QemuOpts *opts = NULL;
4671     const char *fmt = NULL, *src_fmt = NULL, *src_filename = NULL;
4672     const char *filename, *bitmap;
4673     BlockBackend *blk = NULL, *src = NULL;
4674     BlockDriverState *bs = NULL, *src_bs = NULL;
4675     bool image_opts = false;
4676     int64_t granularity = 0;
4677     bool add = false, merge = false;
4678     QSIMPLEQ_HEAD(, ImgBitmapAction) actions;
4679     ImgBitmapAction *act, *act_next;
4680     const char *op;
4681     int inactivate_ret;
4682 
4683     QSIMPLEQ_INIT(&actions);
4684 
4685     for (;;) {
4686         static const struct option long_options[] = {
4687             {"help", no_argument, 0, 'h'},
4688             {"object", required_argument, 0, OPTION_OBJECT},
4689             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4690             {"add", no_argument, 0, OPTION_ADD},
4691             {"remove", no_argument, 0, OPTION_REMOVE},
4692             {"clear", no_argument, 0, OPTION_CLEAR},
4693             {"enable", no_argument, 0, OPTION_ENABLE},
4694             {"disable", no_argument, 0, OPTION_DISABLE},
4695             {"merge", required_argument, 0, OPTION_MERGE},
4696             {"granularity", required_argument, 0, 'g'},
4697             {"source-file", required_argument, 0, 'b'},
4698             {"source-format", required_argument, 0, 'F'},
4699             {0, 0, 0, 0}
4700         };
4701         c = getopt_long(argc, argv, ":b:f:F:g:h", long_options, NULL);
4702         if (c == -1) {
4703             break;
4704         }
4705 
4706         switch (c) {
4707         case ':':
4708             missing_argument(argv[optind - 1]);
4709             break;
4710         case '?':
4711             unrecognized_option(argv[optind - 1]);
4712             break;
4713         case 'h':
4714             help();
4715             break;
4716         case 'b':
4717             src_filename = optarg;
4718             break;
4719         case 'f':
4720             fmt = optarg;
4721             break;
4722         case 'F':
4723             src_fmt = optarg;
4724             break;
4725         case 'g':
4726             granularity = cvtnum("granularity", optarg);
4727             if (granularity < 0) {
4728                 return 1;
4729             }
4730             break;
4731         case OPTION_ADD:
4732             act = g_new0(ImgBitmapAction, 1);
4733             act->act = BITMAP_ADD;
4734             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4735             add = true;
4736             break;
4737         case OPTION_REMOVE:
4738             act = g_new0(ImgBitmapAction, 1);
4739             act->act = BITMAP_REMOVE;
4740             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4741             break;
4742         case OPTION_CLEAR:
4743             act = g_new0(ImgBitmapAction, 1);
4744             act->act = BITMAP_CLEAR;
4745             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4746             break;
4747         case OPTION_ENABLE:
4748             act = g_new0(ImgBitmapAction, 1);
4749             act->act = BITMAP_ENABLE;
4750             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4751             break;
4752         case OPTION_DISABLE:
4753             act = g_new0(ImgBitmapAction, 1);
4754             act->act = BITMAP_DISABLE;
4755             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4756             break;
4757         case OPTION_MERGE:
4758             act = g_new0(ImgBitmapAction, 1);
4759             act->act = BITMAP_MERGE;
4760             act->src = optarg;
4761             QSIMPLEQ_INSERT_TAIL(&actions, act, next);
4762             merge = true;
4763             break;
4764         case OPTION_OBJECT:
4765             user_creatable_process_cmdline(optarg);
4766             break;
4767         case OPTION_IMAGE_OPTS:
4768             image_opts = true;
4769             break;
4770         }
4771     }
4772 
4773     if (QSIMPLEQ_EMPTY(&actions)) {
4774         error_report("Need at least one of --add, --remove, --clear, "
4775                      "--enable, --disable, or --merge");
4776         goto out;
4777     }
4778 
4779     if (granularity && !add) {
4780         error_report("granularity only supported with --add");
4781         goto out;
4782     }
4783     if (src_fmt && !src_filename) {
4784         error_report("-F only supported with -b");
4785         goto out;
4786     }
4787     if (src_filename && !merge) {
4788         error_report("Merge bitmap source file only supported with "
4789                      "--merge");
4790         goto out;
4791     }
4792 
4793     if (optind != argc - 2) {
4794         error_report("Expecting filename and bitmap name");
4795         goto out;
4796     }
4797 
4798     filename = argv[optind];
4799     bitmap = argv[optind + 1];
4800 
4801     /*
4802      * No need to open backing chains; we will be manipulating bitmaps
4803      * directly in this image without reference to image contents.
4804      */
4805     blk = img_open(image_opts, filename, fmt, BDRV_O_RDWR | BDRV_O_NO_BACKING,
4806                    false, false, false);
4807     if (!blk) {
4808         goto out;
4809     }
4810     bs = blk_bs(blk);
4811     if (src_filename) {
4812         src = img_open(false, src_filename, src_fmt, BDRV_O_NO_BACKING,
4813                        false, false, false);
4814         if (!src) {
4815             goto out;
4816         }
4817         src_bs = blk_bs(src);
4818     } else {
4819         src_bs = bs;
4820     }
4821 
4822     QSIMPLEQ_FOREACH_SAFE(act, &actions, next, act_next) {
4823         switch (act->act) {
4824         case BITMAP_ADD:
4825             qmp_block_dirty_bitmap_add(bs->node_name, bitmap,
4826                                        !!granularity, granularity, true, true,
4827                                        false, false, &err);
4828             op = "add";
4829             break;
4830         case BITMAP_REMOVE:
4831             qmp_block_dirty_bitmap_remove(bs->node_name, bitmap, &err);
4832             op = "remove";
4833             break;
4834         case BITMAP_CLEAR:
4835             qmp_block_dirty_bitmap_clear(bs->node_name, bitmap, &err);
4836             op = "clear";
4837             break;
4838         case BITMAP_ENABLE:
4839             qmp_block_dirty_bitmap_enable(bs->node_name, bitmap, &err);
4840             op = "enable";
4841             break;
4842         case BITMAP_DISABLE:
4843             qmp_block_dirty_bitmap_disable(bs->node_name, bitmap, &err);
4844             op = "disable";
4845             break;
4846         case BITMAP_MERGE:
4847             do_dirty_bitmap_merge(bs->node_name, bitmap, src_bs->node_name,
4848                                   act->src, &err);
4849             op = "merge";
4850             break;
4851         default:
4852             g_assert_not_reached();
4853         }
4854 
4855         if (err) {
4856             error_reportf_err(err, "Operation %s on bitmap %s failed: ",
4857                               op, bitmap);
4858             goto out;
4859         }
4860         g_free(act);
4861     }
4862 
4863     ret = 0;
4864 
4865  out:
4866     /*
4867      * Manually inactivate the images first because this way we can know whether
4868      * an error occurred. blk_unref() doesn't tell us about failures.
4869      */
4870     inactivate_ret = bdrv_inactivate_all();
4871     if (inactivate_ret < 0) {
4872         error_report("Error while closing the image: %s", strerror(-inactivate_ret));
4873         ret = 1;
4874     }
4875 
4876     blk_unref(src);
4877     blk_unref(blk);
4878     qemu_opts_del(opts);
4879     return ret;
4880 }
4881 
4882 #define C_BS      01
4883 #define C_COUNT   02
4884 #define C_IF      04
4885 #define C_OF      010
4886 #define C_SKIP    020
4887 
4888 struct DdInfo {
4889     unsigned int flags;
4890     int64_t count;
4891 };
4892 
4893 struct DdIo {
4894     int bsz;    /* Block size */
4895     char *filename;
4896     uint8_t *buf;
4897     int64_t offset;
4898 };
4899 
4900 struct DdOpts {
4901     const char *name;
4902     int (*f)(const char *, struct DdIo *, struct DdIo *, struct DdInfo *);
4903     unsigned int flag;
4904 };
4905 
4906 static int img_dd_bs(const char *arg,
4907                      struct DdIo *in, struct DdIo *out,
4908                      struct DdInfo *dd)
4909 {
4910     int64_t res;
4911 
4912     res = cvtnum_full("bs", arg, 1, INT_MAX);
4913 
4914     if (res < 0) {
4915         return 1;
4916     }
4917     in->bsz = out->bsz = res;
4918 
4919     return 0;
4920 }
4921 
4922 static int img_dd_count(const char *arg,
4923                         struct DdIo *in, struct DdIo *out,
4924                         struct DdInfo *dd)
4925 {
4926     dd->count = cvtnum("count", arg);
4927 
4928     if (dd->count < 0) {
4929         return 1;
4930     }
4931 
4932     return 0;
4933 }
4934 
4935 static int img_dd_if(const char *arg,
4936                      struct DdIo *in, struct DdIo *out,
4937                      struct DdInfo *dd)
4938 {
4939     in->filename = g_strdup(arg);
4940 
4941     return 0;
4942 }
4943 
4944 static int img_dd_of(const char *arg,
4945                      struct DdIo *in, struct DdIo *out,
4946                      struct DdInfo *dd)
4947 {
4948     out->filename = g_strdup(arg);
4949 
4950     return 0;
4951 }
4952 
4953 static int img_dd_skip(const char *arg,
4954                        struct DdIo *in, struct DdIo *out,
4955                        struct DdInfo *dd)
4956 {
4957     in->offset = cvtnum("skip", arg);
4958 
4959     if (in->offset < 0) {
4960         return 1;
4961     }
4962 
4963     return 0;
4964 }
4965 
4966 static int img_dd(int argc, char **argv)
4967 {
4968     int ret = 0;
4969     char *arg = NULL;
4970     char *tmp;
4971     BlockDriver *drv = NULL, *proto_drv = NULL;
4972     BlockBackend *blk1 = NULL, *blk2 = NULL;
4973     QemuOpts *opts = NULL;
4974     QemuOptsList *create_opts = NULL;
4975     Error *local_err = NULL;
4976     bool image_opts = false;
4977     int c, i;
4978     const char *out_fmt = "raw";
4979     const char *fmt = NULL;
4980     int64_t size = 0;
4981     int64_t out_pos, in_pos;
4982     bool force_share = false;
4983     struct DdInfo dd = {
4984         .flags = 0,
4985         .count = 0,
4986     };
4987     struct DdIo in = {
4988         .bsz = 512, /* Block size is by default 512 bytes */
4989         .filename = NULL,
4990         .buf = NULL,
4991         .offset = 0
4992     };
4993     struct DdIo out = {
4994         .bsz = 512,
4995         .filename = NULL,
4996         .buf = NULL,
4997         .offset = 0
4998     };
4999 
5000     const struct DdOpts options[] = {
5001         { "bs", img_dd_bs, C_BS },
5002         { "count", img_dd_count, C_COUNT },
5003         { "if", img_dd_if, C_IF },
5004         { "of", img_dd_of, C_OF },
5005         { "skip", img_dd_skip, C_SKIP },
5006         { NULL, NULL, 0 }
5007     };
5008     const struct option long_options[] = {
5009         { "help", no_argument, 0, 'h'},
5010         { "object", required_argument, 0, OPTION_OBJECT},
5011         { "image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
5012         { "force-share", no_argument, 0, 'U'},
5013         { 0, 0, 0, 0 }
5014     };
5015 
5016     while ((c = getopt_long(argc, argv, ":hf:O:U", long_options, NULL))) {
5017         if (c == EOF) {
5018             break;
5019         }
5020         switch (c) {
5021         case 'O':
5022             out_fmt = optarg;
5023             break;
5024         case 'f':
5025             fmt = optarg;
5026             break;
5027         case ':':
5028             missing_argument(argv[optind - 1]);
5029             break;
5030         case '?':
5031             unrecognized_option(argv[optind - 1]);
5032             break;
5033         case 'h':
5034             help();
5035             break;
5036         case 'U':
5037             force_share = true;
5038             break;
5039         case OPTION_OBJECT:
5040             user_creatable_process_cmdline(optarg);
5041             break;
5042         case OPTION_IMAGE_OPTS:
5043             image_opts = true;
5044             break;
5045         }
5046     }
5047 
5048     for (i = optind; i < argc; i++) {
5049         int j;
5050         arg = g_strdup(argv[i]);
5051 
5052         tmp = strchr(arg, '=');
5053         if (tmp == NULL) {
5054             error_report("unrecognized operand %s", arg);
5055             ret = -1;
5056             goto out;
5057         }
5058 
5059         *tmp++ = '\0';
5060 
5061         for (j = 0; options[j].name != NULL; j++) {
5062             if (!strcmp(arg, options[j].name)) {
5063                 break;
5064             }
5065         }
5066         if (options[j].name == NULL) {
5067             error_report("unrecognized operand %s", arg);
5068             ret = -1;
5069             goto out;
5070         }
5071 
5072         if (options[j].f(tmp, &in, &out, &dd) != 0) {
5073             ret = -1;
5074             goto out;
5075         }
5076         dd.flags |= options[j].flag;
5077         g_free(arg);
5078         arg = NULL;
5079     }
5080 
5081     if (!(dd.flags & C_IF && dd.flags & C_OF)) {
5082         error_report("Must specify both input and output files");
5083         ret = -1;
5084         goto out;
5085     }
5086 
5087     blk1 = img_open(image_opts, in.filename, fmt, 0, false, false,
5088                     force_share);
5089 
5090     if (!blk1) {
5091         ret = -1;
5092         goto out;
5093     }
5094 
5095     drv = bdrv_find_format(out_fmt);
5096     if (!drv) {
5097         error_report("Unknown file format");
5098         ret = -1;
5099         goto out;
5100     }
5101     proto_drv = bdrv_find_protocol(out.filename, true, &local_err);
5102 
5103     if (!proto_drv) {
5104         error_report_err(local_err);
5105         ret = -1;
5106         goto out;
5107     }
5108     if (!drv->create_opts) {
5109         error_report("Format driver '%s' does not support image creation",
5110                      drv->format_name);
5111         ret = -1;
5112         goto out;
5113     }
5114     if (!proto_drv->create_opts) {
5115         error_report("Protocol driver '%s' does not support image creation",
5116                      proto_drv->format_name);
5117         ret = -1;
5118         goto out;
5119     }
5120     create_opts = qemu_opts_append(create_opts, drv->create_opts);
5121     create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5122 
5123     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5124 
5125     size = blk_getlength(blk1);
5126     if (size < 0) {
5127         error_report("Failed to get size for '%s'", in.filename);
5128         ret = -1;
5129         goto out;
5130     }
5131 
5132     if (dd.flags & C_COUNT && dd.count <= INT64_MAX / in.bsz &&
5133         dd.count * in.bsz < size) {
5134         size = dd.count * in.bsz;
5135     }
5136 
5137     /* Overflow means the specified offset is beyond input image's size */
5138     if (dd.flags & C_SKIP && (in.offset > INT64_MAX / in.bsz ||
5139                               size < in.bsz * in.offset)) {
5140         qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 0, &error_abort);
5141     } else {
5142         qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
5143                             size - in.bsz * in.offset, &error_abort);
5144     }
5145 
5146     ret = bdrv_create(drv, out.filename, opts, &local_err);
5147     if (ret < 0) {
5148         error_reportf_err(local_err,
5149                           "%s: error while creating output image: ",
5150                           out.filename);
5151         ret = -1;
5152         goto out;
5153     }
5154 
5155     /* TODO, we can't honour --image-opts for the target,
5156      * since it needs to be given in a format compatible
5157      * with the bdrv_create() call above which does not
5158      * support image-opts style.
5159      */
5160     blk2 = img_open_file(out.filename, NULL, out_fmt, BDRV_O_RDWR,
5161                          false, false, false);
5162 
5163     if (!blk2) {
5164         ret = -1;
5165         goto out;
5166     }
5167 
5168     if (dd.flags & C_SKIP && (in.offset > INT64_MAX / in.bsz ||
5169                               size < in.offset * in.bsz)) {
5170         /* We give a warning if the skip option is bigger than the input
5171          * size and create an empty output disk image (i.e. like dd(1)).
5172          */
5173         error_report("%s: cannot skip to specified offset", in.filename);
5174         in_pos = size;
5175     } else {
5176         in_pos = in.offset * in.bsz;
5177     }
5178 
5179     in.buf = g_new(uint8_t, in.bsz);
5180 
5181     for (out_pos = 0; in_pos < size; ) {
5182         int bytes = (in_pos + in.bsz > size) ? size - in_pos : in.bsz;
5183 
5184         ret = blk_pread(blk1, in_pos, bytes, in.buf, 0);
5185         if (ret < 0) {
5186             error_report("error while reading from input image file: %s",
5187                          strerror(-ret));
5188             goto out;
5189         }
5190         in_pos += bytes;
5191 
5192         ret = blk_pwrite(blk2, out_pos, bytes, in.buf, 0);
5193         if (ret < 0) {
5194             error_report("error while writing to output image file: %s",
5195                          strerror(-ret));
5196             goto out;
5197         }
5198         out_pos += bytes;
5199     }
5200 
5201 out:
5202     g_free(arg);
5203     qemu_opts_del(opts);
5204     qemu_opts_free(create_opts);
5205     blk_unref(blk1);
5206     blk_unref(blk2);
5207     g_free(in.filename);
5208     g_free(out.filename);
5209     g_free(in.buf);
5210     g_free(out.buf);
5211 
5212     if (ret) {
5213         return 1;
5214     }
5215     return 0;
5216 }
5217 
5218 static void dump_json_block_measure_info(BlockMeasureInfo *info)
5219 {
5220     GString *str;
5221     QObject *obj;
5222     Visitor *v = qobject_output_visitor_new(&obj);
5223 
5224     visit_type_BlockMeasureInfo(v, NULL, &info, &error_abort);
5225     visit_complete(v, &obj);
5226     str = qobject_to_json_pretty(obj, true);
5227     assert(str != NULL);
5228     printf("%s\n", str->str);
5229     qobject_unref(obj);
5230     visit_free(v);
5231     g_string_free(str, true);
5232 }
5233 
5234 static int img_measure(int argc, char **argv)
5235 {
5236     static const struct option long_options[] = {
5237         {"help", no_argument, 0, 'h'},
5238         {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
5239         {"object", required_argument, 0, OPTION_OBJECT},
5240         {"output", required_argument, 0, OPTION_OUTPUT},
5241         {"size", required_argument, 0, OPTION_SIZE},
5242         {"force-share", no_argument, 0, 'U'},
5243         {0, 0, 0, 0}
5244     };
5245     OutputFormat output_format = OFORMAT_HUMAN;
5246     BlockBackend *in_blk = NULL;
5247     BlockDriver *drv;
5248     const char *filename = NULL;
5249     const char *fmt = NULL;
5250     const char *out_fmt = "raw";
5251     char *options = NULL;
5252     char *snapshot_name = NULL;
5253     bool force_share = false;
5254     QemuOpts *opts = NULL;
5255     QemuOpts *object_opts = NULL;
5256     QemuOpts *sn_opts = NULL;
5257     QemuOptsList *create_opts = NULL;
5258     bool image_opts = false;
5259     uint64_t img_size = UINT64_MAX;
5260     BlockMeasureInfo *info = NULL;
5261     Error *local_err = NULL;
5262     int ret = 1;
5263     int c;
5264 
5265     while ((c = getopt_long(argc, argv, "hf:O:o:l:U",
5266                             long_options, NULL)) != -1) {
5267         switch (c) {
5268         case '?':
5269         case 'h':
5270             help();
5271             break;
5272         case 'f':
5273             fmt = optarg;
5274             break;
5275         case 'O':
5276             out_fmt = optarg;
5277             break;
5278         case 'o':
5279             if (accumulate_options(&options, optarg) < 0) {
5280                 goto out;
5281             }
5282             break;
5283         case 'l':
5284             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
5285                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
5286                                                   optarg, false);
5287                 if (!sn_opts) {
5288                     error_report("Failed in parsing snapshot param '%s'",
5289                                  optarg);
5290                     goto out;
5291                 }
5292             } else {
5293                 snapshot_name = optarg;
5294             }
5295             break;
5296         case 'U':
5297             force_share = true;
5298             break;
5299         case OPTION_OBJECT:
5300             user_creatable_process_cmdline(optarg);
5301             break;
5302         case OPTION_IMAGE_OPTS:
5303             image_opts = true;
5304             break;
5305         case OPTION_OUTPUT:
5306             if (!strcmp(optarg, "json")) {
5307                 output_format = OFORMAT_JSON;
5308             } else if (!strcmp(optarg, "human")) {
5309                 output_format = OFORMAT_HUMAN;
5310             } else {
5311                 error_report("--output must be used with human or json "
5312                              "as argument.");
5313                 goto out;
5314             }
5315             break;
5316         case OPTION_SIZE:
5317         {
5318             int64_t sval;
5319 
5320             sval = cvtnum("image size", optarg);
5321             if (sval < 0) {
5322                 goto out;
5323             }
5324             img_size = (uint64_t)sval;
5325         }
5326         break;
5327         }
5328     }
5329 
5330     if (argc - optind > 1) {
5331         error_report("At most one filename argument is allowed.");
5332         goto out;
5333     } else if (argc - optind == 1) {
5334         filename = argv[optind];
5335     }
5336 
5337     if (!filename && (image_opts || fmt || snapshot_name || sn_opts)) {
5338         error_report("--image-opts, -f, and -l require a filename argument.");
5339         goto out;
5340     }
5341     if (filename && img_size != UINT64_MAX) {
5342         error_report("--size N cannot be used together with a filename.");
5343         goto out;
5344     }
5345     if (!filename && img_size == UINT64_MAX) {
5346         error_report("Either --size N or one filename must be specified.");
5347         goto out;
5348     }
5349 
5350     if (filename) {
5351         in_blk = img_open(image_opts, filename, fmt, 0,
5352                           false, false, force_share);
5353         if (!in_blk) {
5354             goto out;
5355         }
5356 
5357         if (sn_opts) {
5358             bdrv_snapshot_load_tmp(blk_bs(in_blk),
5359                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
5360                     qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
5361                     &local_err);
5362         } else if (snapshot_name != NULL) {
5363             bdrv_snapshot_load_tmp_by_id_or_name(blk_bs(in_blk),
5364                     snapshot_name, &local_err);
5365         }
5366         if (local_err) {
5367             error_reportf_err(local_err, "Failed to load snapshot: ");
5368             goto out;
5369         }
5370     }
5371 
5372     drv = bdrv_find_format(out_fmt);
5373     if (!drv) {
5374         error_report("Unknown file format '%s'", out_fmt);
5375         goto out;
5376     }
5377     if (!drv->create_opts) {
5378         error_report("Format driver '%s' does not support image creation",
5379                      drv->format_name);
5380         goto out;
5381     }
5382 
5383     create_opts = qemu_opts_append(create_opts, drv->create_opts);
5384     create_opts = qemu_opts_append(create_opts, bdrv_file.create_opts);
5385     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5386     if (options) {
5387         if (!qemu_opts_do_parse(opts, options, NULL, &local_err)) {
5388             error_report_err(local_err);
5389             error_report("Invalid options for file format '%s'", out_fmt);
5390             goto out;
5391         }
5392     }
5393     if (img_size != UINT64_MAX) {
5394         qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
5395     }
5396 
5397     info = bdrv_measure(drv, opts, in_blk ? blk_bs(in_blk) : NULL, &local_err);
5398     if (local_err) {
5399         error_report_err(local_err);
5400         goto out;
5401     }
5402 
5403     if (output_format == OFORMAT_HUMAN) {
5404         printf("required size: %" PRIu64 "\n", info->required);
5405         printf("fully allocated size: %" PRIu64 "\n", info->fully_allocated);
5406         if (info->has_bitmaps) {
5407             printf("bitmaps size: %" PRIu64 "\n", info->bitmaps);
5408         }
5409     } else {
5410         dump_json_block_measure_info(info);
5411     }
5412 
5413     ret = 0;
5414 
5415 out:
5416     qapi_free_BlockMeasureInfo(info);
5417     qemu_opts_del(object_opts);
5418     qemu_opts_del(opts);
5419     qemu_opts_del(sn_opts);
5420     qemu_opts_free(create_opts);
5421     g_free(options);
5422     blk_unref(in_blk);
5423     return ret;
5424 }
5425 
5426 static const img_cmd_t img_cmds[] = {
5427 #define DEF(option, callback, arg_string)        \
5428     { option, callback },
5429 #include "qemu-img-cmds.h"
5430 #undef DEF
5431     { NULL, NULL, },
5432 };
5433 
5434 int main(int argc, char **argv)
5435 {
5436     const img_cmd_t *cmd;
5437     const char *cmdname;
5438     int c;
5439     static const struct option long_options[] = {
5440         {"help", no_argument, 0, 'h'},
5441         {"version", no_argument, 0, 'V'},
5442         {"trace", required_argument, NULL, 'T'},
5443         {0, 0, 0, 0}
5444     };
5445 
5446 #ifdef CONFIG_POSIX
5447     signal(SIGPIPE, SIG_IGN);
5448 #endif
5449 
5450     socket_init();
5451     error_init(argv[0]);
5452     module_call_init(MODULE_INIT_TRACE);
5453     qemu_init_exec_dir(argv[0]);
5454 
5455     qemu_init_main_loop(&error_fatal);
5456 
5457     qcrypto_init(&error_fatal);
5458 
5459     module_call_init(MODULE_INIT_QOM);
5460     bdrv_init();
5461     if (argc < 2) {
5462         error_exit("Not enough arguments");
5463     }
5464 
5465     qemu_add_opts(&qemu_source_opts);
5466     qemu_add_opts(&qemu_trace_opts);
5467 
5468     while ((c = getopt_long(argc, argv, "+:hVT:", long_options, NULL)) != -1) {
5469         switch (c) {
5470         case ':':
5471             missing_argument(argv[optind - 1]);
5472             return 0;
5473         case '?':
5474             unrecognized_option(argv[optind - 1]);
5475             return 0;
5476         case 'h':
5477             help();
5478             return 0;
5479         case 'V':
5480             printf(QEMU_IMG_VERSION);
5481             return 0;
5482         case 'T':
5483             trace_opt_parse(optarg);
5484             break;
5485         }
5486     }
5487 
5488     cmdname = argv[optind];
5489 
5490     /* reset getopt_long scanning */
5491     argc -= optind;
5492     if (argc < 1) {
5493         return 0;
5494     }
5495     argv += optind;
5496     qemu_reset_optind();
5497 
5498     if (!trace_init_backends()) {
5499         exit(1);
5500     }
5501     trace_init_file();
5502     qemu_set_log(LOG_TRACE, &error_fatal);
5503 
5504     /* find the command */
5505     for (cmd = img_cmds; cmd->name != NULL; cmd++) {
5506         if (!strcmp(cmdname, cmd->name)) {
5507             return cmd->handler(argc, argv);
5508         }
5509     }
5510 
5511     /* not found */
5512     error_exit("Command not found: %s", cmdname);
5513 }
5514