xref: /qemu/qemu-img.c (revision bf8d4924)
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 #include "qemu/osdep.h"
25 #include "qemu-version.h"
26 #include "qapi/error.h"
27 #include "qapi-visit.h"
28 #include "qapi/qmp-output-visitor.h"
29 #include "qapi/qmp/qerror.h"
30 #include "qapi/qmp/qjson.h"
31 #include "qemu/cutils.h"
32 #include "qemu/config-file.h"
33 #include "qemu/option.h"
34 #include "qemu/error-report.h"
35 #include "qemu/log.h"
36 #include "qom/object_interfaces.h"
37 #include "sysemu/sysemu.h"
38 #include "sysemu/block-backend.h"
39 #include "block/block_int.h"
40 #include "block/blockjob.h"
41 #include "block/qapi.h"
42 #include "crypto/init.h"
43 #include "trace/control.h"
44 #include <getopt.h>
45 
46 #define QEMU_IMG_VERSION "qemu-img version " QEMU_VERSION QEMU_PKGVERSION \
47                           ", Copyright (c) 2004-2008 Fabrice Bellard\n"
48 
49 typedef struct img_cmd_t {
50     const char *name;
51     int (*handler)(int argc, char **argv);
52 } img_cmd_t;
53 
54 enum {
55     OPTION_OUTPUT = 256,
56     OPTION_BACKING_CHAIN = 257,
57     OPTION_OBJECT = 258,
58     OPTION_IMAGE_OPTS = 259,
59     OPTION_PATTERN = 260,
60     OPTION_FLUSH_INTERVAL = 261,
61     OPTION_NO_DRAIN = 262,
62 };
63 
64 typedef enum OutputFormat {
65     OFORMAT_JSON,
66     OFORMAT_HUMAN,
67 } OutputFormat;
68 
69 /* Default to cache=writeback as data integrity is not important for qemu-img */
70 #define BDRV_DEFAULT_CACHE "writeback"
71 
72 static void format_print(void *opaque, const char *name)
73 {
74     printf(" %s", name);
75 }
76 
77 static void QEMU_NORETURN GCC_FMT_ATTR(1, 2) error_exit(const char *fmt, ...)
78 {
79     va_list ap;
80 
81     error_printf("qemu-img: ");
82 
83     va_start(ap, fmt);
84     error_vprintf(fmt, ap);
85     va_end(ap);
86 
87     error_printf("\nTry 'qemu-img --help' for more information\n");
88     exit(EXIT_FAILURE);
89 }
90 
91 /* Please keep in synch with qemu-img.texi */
92 static void QEMU_NORETURN help(void)
93 {
94     const char *help_msg =
95            QEMU_IMG_VERSION
96            "usage: qemu-img [standard options] command [command options]\n"
97            "QEMU disk image utility\n"
98            "\n"
99            "    '-h', '--help'       display this help and exit\n"
100            "    '-V', '--version'    output version information and exit\n"
101            "    '-T', '--trace'      [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
102            "                         specify tracing options\n"
103            "\n"
104            "Command syntax:\n"
105 #define DEF(option, callback, arg_string)        \
106            "  " arg_string "\n"
107 #include "qemu-img-cmds.h"
108 #undef DEF
109 #undef GEN_DOCS
110            "\n"
111            "Command parameters:\n"
112            "  'filename' is a disk image filename\n"
113            "  'objectdef' is a QEMU user creatable object definition. See the qemu(1)\n"
114            "    manual page for a description of the object properties. The most common\n"
115            "    object type is a 'secret', which is used to supply passwords and/or\n"
116            "    encryption keys.\n"
117            "  'fmt' is the disk image format. It is guessed automatically in most cases\n"
118            "  'cache' is the cache mode used to write the output disk image, the valid\n"
119            "    options are: 'none', 'writeback' (default, except for convert), 'writethrough',\n"
120            "    'directsync' and 'unsafe' (default for convert)\n"
121            "  'src_cache' is the cache mode used to read input disk images, the valid\n"
122            "    options are the same as for the 'cache' option\n"
123            "  'size' is the disk image size in bytes. Optional suffixes\n"
124            "    'k' or 'K' (kilobyte, 1024), 'M' (megabyte, 1024k), 'G' (gigabyte, 1024M),\n"
125            "    'T' (terabyte, 1024G), 'P' (petabyte, 1024T) and 'E' (exabyte, 1024P)  are\n"
126            "    supported. 'b' is ignored.\n"
127            "  'output_filename' is the destination disk image filename\n"
128            "  'output_fmt' is the destination format\n"
129            "  'options' is a comma separated list of format specific options in a\n"
130            "    name=value format. Use -o ? for an overview of the options supported by the\n"
131            "    used format\n"
132            "  'snapshot_param' is param used for internal snapshot, format\n"
133            "    is 'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
134            "    '[ID_OR_NAME]'\n"
135            "  'snapshot_id_or_name' is deprecated, use 'snapshot_param'\n"
136            "    instead\n"
137            "  '-c' indicates that target image must be compressed (qcow format only)\n"
138            "  '-u' enables unsafe rebasing. It is assumed that old and new backing file\n"
139            "       match exactly. The image doesn't need a working backing file before\n"
140            "       rebasing in this case (useful for renaming the backing file)\n"
141            "  '-h' with or without a command shows this help and lists the supported formats\n"
142            "  '-p' show progress of command (only certain commands)\n"
143            "  '-q' use Quiet mode - do not print any output (except errors)\n"
144            "  '-S' indicates the consecutive number of bytes (defaults to 4k) that must\n"
145            "       contain only zeros for qemu-img to create a sparse image during\n"
146            "       conversion. If the number of bytes is 0, the source will not be scanned for\n"
147            "       unallocated or zero sectors, and the destination image will always be\n"
148            "       fully allocated\n"
149            "  '--output' takes the format in which the output must be done (human or json)\n"
150            "  '-n' skips the target volume creation (useful if the volume is created\n"
151            "       prior to running qemu-img)\n"
152            "\n"
153            "Parameters to check subcommand:\n"
154            "  '-r' tries to repair any inconsistencies that are found during the check.\n"
155            "       '-r leaks' repairs only cluster leaks, whereas '-r all' fixes all\n"
156            "       kinds of errors, with a higher risk of choosing the wrong fix or\n"
157            "       hiding corruption that has already occurred.\n"
158            "\n"
159            "Parameters to snapshot subcommand:\n"
160            "  'snapshot' is the name of the snapshot to create, apply or delete\n"
161            "  '-a' applies a snapshot (revert disk to saved state)\n"
162            "  '-c' creates a snapshot\n"
163            "  '-d' deletes a snapshot\n"
164            "  '-l' lists all snapshots in the given image\n"
165            "\n"
166            "Parameters to compare subcommand:\n"
167            "  '-f' first image format\n"
168            "  '-F' second image format\n"
169            "  '-s' run in Strict mode - fail on different image size or sector allocation\n";
170 
171     printf("%s\nSupported formats:", help_msg);
172     bdrv_iterate_format(format_print, NULL);
173     printf("\n");
174     exit(EXIT_SUCCESS);
175 }
176 
177 static QemuOptsList qemu_object_opts = {
178     .name = "object",
179     .implied_opt_name = "qom-type",
180     .head = QTAILQ_HEAD_INITIALIZER(qemu_object_opts.head),
181     .desc = {
182         { }
183     },
184 };
185 
186 static QemuOptsList qemu_source_opts = {
187     .name = "source",
188     .implied_opt_name = "file",
189     .head = QTAILQ_HEAD_INITIALIZER(qemu_source_opts.head),
190     .desc = {
191         { }
192     },
193 };
194 
195 static int GCC_FMT_ATTR(2, 3) qprintf(bool quiet, const char *fmt, ...)
196 {
197     int ret = 0;
198     if (!quiet) {
199         va_list args;
200         va_start(args, fmt);
201         ret = vprintf(fmt, args);
202         va_end(args);
203     }
204     return ret;
205 }
206 
207 
208 static int print_block_option_help(const char *filename, const char *fmt)
209 {
210     BlockDriver *drv, *proto_drv;
211     QemuOptsList *create_opts = NULL;
212     Error *local_err = NULL;
213 
214     /* Find driver and parse its options */
215     drv = bdrv_find_format(fmt);
216     if (!drv) {
217         error_report("Unknown file format '%s'", fmt);
218         return 1;
219     }
220 
221     create_opts = qemu_opts_append(create_opts, drv->create_opts);
222     if (filename) {
223         proto_drv = bdrv_find_protocol(filename, true, &local_err);
224         if (!proto_drv) {
225             error_report_err(local_err);
226             qemu_opts_free(create_opts);
227             return 1;
228         }
229         create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
230     }
231 
232     qemu_opts_print_help(create_opts);
233     qemu_opts_free(create_opts);
234     return 0;
235 }
236 
237 
238 static int img_open_password(BlockBackend *blk, const char *filename,
239                              int flags, bool quiet)
240 {
241     BlockDriverState *bs;
242     char password[256];
243 
244     bs = blk_bs(blk);
245     if (bdrv_is_encrypted(bs) && bdrv_key_required(bs) &&
246         !(flags & BDRV_O_NO_IO)) {
247         qprintf(quiet, "Disk image '%s' is encrypted.\n", filename);
248         if (qemu_read_password(password, sizeof(password)) < 0) {
249             error_report("No password given");
250             return -1;
251         }
252         if (bdrv_set_key(bs, password) < 0) {
253             error_report("invalid password");
254             return -1;
255         }
256     }
257     return 0;
258 }
259 
260 
261 static BlockBackend *img_open_opts(const char *optstr,
262                                    QemuOpts *opts, int flags, bool writethrough,
263                                    bool quiet)
264 {
265     QDict *options;
266     Error *local_err = NULL;
267     BlockBackend *blk;
268     options = qemu_opts_to_qdict(opts, NULL);
269     blk = blk_new_open(NULL, NULL, options, flags, &local_err);
270     if (!blk) {
271         error_reportf_err(local_err, "Could not open '%s': ", optstr);
272         return NULL;
273     }
274     blk_set_enable_write_cache(blk, !writethrough);
275 
276     if (img_open_password(blk, optstr, flags, quiet) < 0) {
277         blk_unref(blk);
278         return NULL;
279     }
280     return blk;
281 }
282 
283 static BlockBackend *img_open_file(const char *filename,
284                                    const char *fmt, int flags,
285                                    bool writethrough, bool quiet)
286 {
287     BlockBackend *blk;
288     Error *local_err = NULL;
289     QDict *options = NULL;
290 
291     if (fmt) {
292         options = qdict_new();
293         qdict_put(options, "driver", qstring_from_str(fmt));
294     }
295 
296     blk = blk_new_open(filename, NULL, options, flags, &local_err);
297     if (!blk) {
298         error_reportf_err(local_err, "Could not open '%s': ", filename);
299         return NULL;
300     }
301     blk_set_enable_write_cache(blk, !writethrough);
302 
303     if (img_open_password(blk, filename, flags, quiet) < 0) {
304         blk_unref(blk);
305         return NULL;
306     }
307     return blk;
308 }
309 
310 
311 static BlockBackend *img_open(bool image_opts,
312                               const char *filename,
313                               const char *fmt, int flags, bool writethrough,
314                               bool quiet)
315 {
316     BlockBackend *blk;
317     if (image_opts) {
318         QemuOpts *opts;
319         if (fmt) {
320             error_report("--image-opts and --format are mutually exclusive");
321             return NULL;
322         }
323         opts = qemu_opts_parse_noisily(qemu_find_opts("source"),
324                                        filename, true);
325         if (!opts) {
326             return NULL;
327         }
328         blk = img_open_opts(filename, opts, flags, writethrough, quiet);
329     } else {
330         blk = img_open_file(filename, fmt, flags, writethrough, quiet);
331     }
332     return blk;
333 }
334 
335 
336 static int add_old_style_options(const char *fmt, QemuOpts *opts,
337                                  const char *base_filename,
338                                  const char *base_fmt)
339 {
340     Error *err = NULL;
341 
342     if (base_filename) {
343         qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &err);
344         if (err) {
345             error_report("Backing file not supported for file format '%s'",
346                          fmt);
347             error_free(err);
348             return -1;
349         }
350     }
351     if (base_fmt) {
352         qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &err);
353         if (err) {
354             error_report("Backing file format not supported for file "
355                          "format '%s'", fmt);
356             error_free(err);
357             return -1;
358         }
359     }
360     return 0;
361 }
362 
363 static int img_create(int argc, char **argv)
364 {
365     int c;
366     uint64_t img_size = -1;
367     const char *fmt = "raw";
368     const char *base_fmt = NULL;
369     const char *filename;
370     const char *base_filename = NULL;
371     char *options = NULL;
372     Error *local_err = NULL;
373     bool quiet = false;
374 
375     for(;;) {
376         static const struct option long_options[] = {
377             {"help", no_argument, 0, 'h'},
378             {"object", required_argument, 0, OPTION_OBJECT},
379             {0, 0, 0, 0}
380         };
381         c = getopt_long(argc, argv, "F:b:f:he6o:q",
382                         long_options, NULL);
383         if (c == -1) {
384             break;
385         }
386         switch(c) {
387         case '?':
388         case 'h':
389             help();
390             break;
391         case 'F':
392             base_fmt = optarg;
393             break;
394         case 'b':
395             base_filename = optarg;
396             break;
397         case 'f':
398             fmt = optarg;
399             break;
400         case 'e':
401             error_report("option -e is deprecated, please use \'-o "
402                   "encryption\' instead!");
403             goto fail;
404         case '6':
405             error_report("option -6 is deprecated, please use \'-o "
406                   "compat6\' instead!");
407             goto fail;
408         case 'o':
409             if (!is_valid_option_list(optarg)) {
410                 error_report("Invalid option list: %s", optarg);
411                 goto fail;
412             }
413             if (!options) {
414                 options = g_strdup(optarg);
415             } else {
416                 char *old_options = options;
417                 options = g_strdup_printf("%s,%s", options, optarg);
418                 g_free(old_options);
419             }
420             break;
421         case 'q':
422             quiet = true;
423             break;
424         case OPTION_OBJECT: {
425             QemuOpts *opts;
426             opts = qemu_opts_parse_noisily(&qemu_object_opts,
427                                            optarg, true);
428             if (!opts) {
429                 goto fail;
430             }
431         }   break;
432         }
433     }
434 
435     /* Get the filename */
436     filename = (optind < argc) ? argv[optind] : NULL;
437     if (options && has_help_option(options)) {
438         g_free(options);
439         return print_block_option_help(filename, fmt);
440     }
441 
442     if (optind >= argc) {
443         error_exit("Expecting image file name");
444     }
445     optind++;
446 
447     if (qemu_opts_foreach(&qemu_object_opts,
448                           user_creatable_add_opts_foreach,
449                           NULL, NULL)) {
450         goto fail;
451     }
452 
453     /* Get image size, if specified */
454     if (optind < argc) {
455         int64_t sval;
456         char *end;
457         sval = qemu_strtosz_suffix(argv[optind++], &end,
458                                    QEMU_STRTOSZ_DEFSUFFIX_B);
459         if (sval < 0 || *end) {
460             if (sval == -ERANGE) {
461                 error_report("Image size must be less than 8 EiB!");
462             } else {
463                 error_report("Invalid image size specified! You may use k, M, "
464                       "G, T, P or E suffixes for ");
465                 error_report("kilobytes, megabytes, gigabytes, terabytes, "
466                              "petabytes and exabytes.");
467             }
468             goto fail;
469         }
470         img_size = (uint64_t)sval;
471     }
472     if (optind != argc) {
473         error_exit("Unexpected argument: %s", argv[optind]);
474     }
475 
476     bdrv_img_create(filename, fmt, base_filename, base_fmt,
477                     options, img_size, 0, &local_err, quiet);
478     if (local_err) {
479         error_reportf_err(local_err, "%s: ", filename);
480         goto fail;
481     }
482 
483     g_free(options);
484     return 0;
485 
486 fail:
487     g_free(options);
488     return 1;
489 }
490 
491 static void dump_json_image_check(ImageCheck *check, bool quiet)
492 {
493     Error *local_err = NULL;
494     QString *str;
495     QmpOutputVisitor *ov = qmp_output_visitor_new();
496     QObject *obj;
497     visit_type_ImageCheck(qmp_output_get_visitor(ov), NULL, &check,
498                           &local_err);
499     obj = qmp_output_get_qobject(ov);
500     str = qobject_to_json_pretty(obj);
501     assert(str != NULL);
502     qprintf(quiet, "%s\n", qstring_get_str(str));
503     qobject_decref(obj);
504     qmp_output_visitor_cleanup(ov);
505     QDECREF(str);
506 }
507 
508 static void dump_human_image_check(ImageCheck *check, bool quiet)
509 {
510     if (!(check->corruptions || check->leaks || check->check_errors)) {
511         qprintf(quiet, "No errors were found on the image.\n");
512     } else {
513         if (check->corruptions) {
514             qprintf(quiet, "\n%" PRId64 " errors were found on the image.\n"
515                     "Data may be corrupted, or further writes to the image "
516                     "may corrupt it.\n",
517                     check->corruptions);
518         }
519 
520         if (check->leaks) {
521             qprintf(quiet,
522                     "\n%" PRId64 " leaked clusters were found on the image.\n"
523                     "This means waste of disk space, but no harm to data.\n",
524                     check->leaks);
525         }
526 
527         if (check->check_errors) {
528             qprintf(quiet,
529                     "\n%" PRId64
530                     " internal errors have occurred during the check.\n",
531                     check->check_errors);
532         }
533     }
534 
535     if (check->total_clusters != 0 && check->allocated_clusters != 0) {
536         qprintf(quiet, "%" PRId64 "/%" PRId64 " = %0.2f%% allocated, "
537                 "%0.2f%% fragmented, %0.2f%% compressed clusters\n",
538                 check->allocated_clusters, check->total_clusters,
539                 check->allocated_clusters * 100.0 / check->total_clusters,
540                 check->fragmented_clusters * 100.0 / check->allocated_clusters,
541                 check->compressed_clusters * 100.0 /
542                 check->allocated_clusters);
543     }
544 
545     if (check->image_end_offset) {
546         qprintf(quiet,
547                 "Image end offset: %" PRId64 "\n", check->image_end_offset);
548     }
549 }
550 
551 static int collect_image_check(BlockDriverState *bs,
552                    ImageCheck *check,
553                    const char *filename,
554                    const char *fmt,
555                    int fix)
556 {
557     int ret;
558     BdrvCheckResult result;
559 
560     ret = bdrv_check(bs, &result, fix);
561     if (ret < 0) {
562         return ret;
563     }
564 
565     check->filename                 = g_strdup(filename);
566     check->format                   = g_strdup(bdrv_get_format_name(bs));
567     check->check_errors             = result.check_errors;
568     check->corruptions              = result.corruptions;
569     check->has_corruptions          = result.corruptions != 0;
570     check->leaks                    = result.leaks;
571     check->has_leaks                = result.leaks != 0;
572     check->corruptions_fixed        = result.corruptions_fixed;
573     check->has_corruptions_fixed    = result.corruptions != 0;
574     check->leaks_fixed              = result.leaks_fixed;
575     check->has_leaks_fixed          = result.leaks != 0;
576     check->image_end_offset         = result.image_end_offset;
577     check->has_image_end_offset     = result.image_end_offset != 0;
578     check->total_clusters           = result.bfi.total_clusters;
579     check->has_total_clusters       = result.bfi.total_clusters != 0;
580     check->allocated_clusters       = result.bfi.allocated_clusters;
581     check->has_allocated_clusters   = result.bfi.allocated_clusters != 0;
582     check->fragmented_clusters      = result.bfi.fragmented_clusters;
583     check->has_fragmented_clusters  = result.bfi.fragmented_clusters != 0;
584     check->compressed_clusters      = result.bfi.compressed_clusters;
585     check->has_compressed_clusters  = result.bfi.compressed_clusters != 0;
586 
587     return 0;
588 }
589 
590 /*
591  * Checks an image for consistency. Exit codes:
592  *
593  *  0 - Check completed, image is good
594  *  1 - Check not completed because of internal errors
595  *  2 - Check completed, image is corrupted
596  *  3 - Check completed, image has leaked clusters, but is good otherwise
597  * 63 - Checks are not supported by the image format
598  */
599 static int img_check(int argc, char **argv)
600 {
601     int c, ret;
602     OutputFormat output_format = OFORMAT_HUMAN;
603     const char *filename, *fmt, *output, *cache;
604     BlockBackend *blk;
605     BlockDriverState *bs;
606     int fix = 0;
607     int flags = BDRV_O_CHECK;
608     bool writethrough;
609     ImageCheck *check;
610     bool quiet = false;
611     bool image_opts = false;
612 
613     fmt = NULL;
614     output = NULL;
615     cache = BDRV_DEFAULT_CACHE;
616 
617     for(;;) {
618         int option_index = 0;
619         static const struct option long_options[] = {
620             {"help", no_argument, 0, 'h'},
621             {"format", required_argument, 0, 'f'},
622             {"repair", required_argument, 0, 'r'},
623             {"output", required_argument, 0, OPTION_OUTPUT},
624             {"object", required_argument, 0, OPTION_OBJECT},
625             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
626             {0, 0, 0, 0}
627         };
628         c = getopt_long(argc, argv, "hf:r:T:q",
629                         long_options, &option_index);
630         if (c == -1) {
631             break;
632         }
633         switch(c) {
634         case '?':
635         case 'h':
636             help();
637             break;
638         case 'f':
639             fmt = optarg;
640             break;
641         case 'r':
642             flags |= BDRV_O_RDWR;
643 
644             if (!strcmp(optarg, "leaks")) {
645                 fix = BDRV_FIX_LEAKS;
646             } else if (!strcmp(optarg, "all")) {
647                 fix = BDRV_FIX_LEAKS | BDRV_FIX_ERRORS;
648             } else {
649                 error_exit("Unknown option value for -r "
650                            "(expecting 'leaks' or 'all'): %s", optarg);
651             }
652             break;
653         case OPTION_OUTPUT:
654             output = optarg;
655             break;
656         case 'T':
657             cache = optarg;
658             break;
659         case 'q':
660             quiet = true;
661             break;
662         case OPTION_OBJECT: {
663             QemuOpts *opts;
664             opts = qemu_opts_parse_noisily(&qemu_object_opts,
665                                            optarg, true);
666             if (!opts) {
667                 return 1;
668             }
669         }   break;
670         case OPTION_IMAGE_OPTS:
671             image_opts = true;
672             break;
673         }
674     }
675     if (optind != argc - 1) {
676         error_exit("Expecting one image file name");
677     }
678     filename = argv[optind++];
679 
680     if (output && !strcmp(output, "json")) {
681         output_format = OFORMAT_JSON;
682     } else if (output && !strcmp(output, "human")) {
683         output_format = OFORMAT_HUMAN;
684     } else if (output) {
685         error_report("--output must be used with human or json as argument.");
686         return 1;
687     }
688 
689     if (qemu_opts_foreach(&qemu_object_opts,
690                           user_creatable_add_opts_foreach,
691                           NULL, NULL)) {
692         return 1;
693     }
694 
695     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
696     if (ret < 0) {
697         error_report("Invalid source cache option: %s", cache);
698         return 1;
699     }
700 
701     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet);
702     if (!blk) {
703         return 1;
704     }
705     bs = blk_bs(blk);
706 
707     check = g_new0(ImageCheck, 1);
708     ret = collect_image_check(bs, check, filename, fmt, fix);
709 
710     if (ret == -ENOTSUP) {
711         error_report("This image format does not support checks");
712         ret = 63;
713         goto fail;
714     }
715 
716     if (check->corruptions_fixed || check->leaks_fixed) {
717         int corruptions_fixed, leaks_fixed;
718 
719         leaks_fixed         = check->leaks_fixed;
720         corruptions_fixed   = check->corruptions_fixed;
721 
722         if (output_format == OFORMAT_HUMAN) {
723             qprintf(quiet,
724                     "The following inconsistencies were found and repaired:\n\n"
725                     "    %" PRId64 " leaked clusters\n"
726                     "    %" PRId64 " corruptions\n\n"
727                     "Double checking the fixed image now...\n",
728                     check->leaks_fixed,
729                     check->corruptions_fixed);
730         }
731 
732         ret = collect_image_check(bs, check, filename, fmt, 0);
733 
734         check->leaks_fixed          = leaks_fixed;
735         check->corruptions_fixed    = corruptions_fixed;
736     }
737 
738     if (!ret) {
739         switch (output_format) {
740         case OFORMAT_HUMAN:
741             dump_human_image_check(check, quiet);
742             break;
743         case OFORMAT_JSON:
744             dump_json_image_check(check, quiet);
745             break;
746         }
747     }
748 
749     if (ret || check->check_errors) {
750         if (ret) {
751             error_report("Check failed: %s", strerror(-ret));
752         } else {
753             error_report("Check failed");
754         }
755         ret = 1;
756         goto fail;
757     }
758 
759     if (check->corruptions) {
760         ret = 2;
761     } else if (check->leaks) {
762         ret = 3;
763     } else {
764         ret = 0;
765     }
766 
767 fail:
768     qapi_free_ImageCheck(check);
769     blk_unref(blk);
770     return ret;
771 }
772 
773 typedef struct CommonBlockJobCBInfo {
774     BlockDriverState *bs;
775     Error **errp;
776 } CommonBlockJobCBInfo;
777 
778 static void common_block_job_cb(void *opaque, int ret)
779 {
780     CommonBlockJobCBInfo *cbi = opaque;
781 
782     if (ret < 0) {
783         error_setg_errno(cbi->errp, -ret, "Block job failed");
784     }
785 }
786 
787 static void run_block_job(BlockJob *job, Error **errp)
788 {
789     AioContext *aio_context = blk_get_aio_context(job->blk);
790 
791     do {
792         aio_poll(aio_context, true);
793         qemu_progress_print(job->len ?
794                             ((float)job->offset / job->len * 100.f) : 0.0f, 0);
795     } while (!job->ready);
796 
797     block_job_complete_sync(job, errp);
798 
799     /* A block job may finish instantaneously without publishing any progress,
800      * so just signal completion here */
801     qemu_progress_print(100.f, 0);
802 }
803 
804 static int img_commit(int argc, char **argv)
805 {
806     int c, ret, flags;
807     const char *filename, *fmt, *cache, *base;
808     BlockBackend *blk;
809     BlockDriverState *bs, *base_bs;
810     bool progress = false, quiet = false, drop = false;
811     bool writethrough;
812     Error *local_err = NULL;
813     CommonBlockJobCBInfo cbi;
814     bool image_opts = false;
815 
816     fmt = NULL;
817     cache = BDRV_DEFAULT_CACHE;
818     base = NULL;
819     for(;;) {
820         static const struct option long_options[] = {
821             {"help", no_argument, 0, 'h'},
822             {"object", required_argument, 0, OPTION_OBJECT},
823             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
824             {0, 0, 0, 0}
825         };
826         c = getopt_long(argc, argv, "f:ht:b:dpq",
827                         long_options, NULL);
828         if (c == -1) {
829             break;
830         }
831         switch(c) {
832         case '?':
833         case 'h':
834             help();
835             break;
836         case 'f':
837             fmt = optarg;
838             break;
839         case 't':
840             cache = optarg;
841             break;
842         case 'b':
843             base = optarg;
844             /* -b implies -d */
845             drop = true;
846             break;
847         case 'd':
848             drop = true;
849             break;
850         case 'p':
851             progress = true;
852             break;
853         case 'q':
854             quiet = true;
855             break;
856         case OPTION_OBJECT: {
857             QemuOpts *opts;
858             opts = qemu_opts_parse_noisily(&qemu_object_opts,
859                                            optarg, true);
860             if (!opts) {
861                 return 1;
862             }
863         }   break;
864         case OPTION_IMAGE_OPTS:
865             image_opts = true;
866             break;
867         }
868     }
869 
870     /* Progress is not shown in Quiet mode */
871     if (quiet) {
872         progress = false;
873     }
874 
875     if (optind != argc - 1) {
876         error_exit("Expecting one image file name");
877     }
878     filename = argv[optind++];
879 
880     if (qemu_opts_foreach(&qemu_object_opts,
881                           user_creatable_add_opts_foreach,
882                           NULL, NULL)) {
883         return 1;
884     }
885 
886     flags = BDRV_O_RDWR | BDRV_O_UNMAP;
887     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
888     if (ret < 0) {
889         error_report("Invalid cache option: %s", cache);
890         return 1;
891     }
892 
893     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet);
894     if (!blk) {
895         return 1;
896     }
897     bs = blk_bs(blk);
898 
899     qemu_progress_init(progress, 1.f);
900     qemu_progress_print(0.f, 100);
901 
902     if (base) {
903         base_bs = bdrv_find_backing_image(bs, base);
904         if (!base_bs) {
905             error_setg(&local_err, QERR_BASE_NOT_FOUND, base);
906             goto done;
907         }
908     } else {
909         /* This is different from QMP, which by default uses the deepest file in
910          * the backing chain (i.e., the very base); however, the traditional
911          * behavior of qemu-img commit is using the immediate backing file. */
912         base_bs = backing_bs(bs);
913         if (!base_bs) {
914             error_setg(&local_err, "Image does not have a backing file");
915             goto done;
916         }
917     }
918 
919     cbi = (CommonBlockJobCBInfo){
920         .errp = &local_err,
921         .bs   = bs,
922     };
923 
924     commit_active_start(bs, base_bs, 0, BLOCKDEV_ON_ERROR_REPORT,
925                         common_block_job_cb, &cbi, &local_err);
926     if (local_err) {
927         goto done;
928     }
929 
930     /* When the block job completes, the BlockBackend reference will point to
931      * the old backing file. In order to avoid that the top image is already
932      * deleted, so we can still empty it afterwards, increment the reference
933      * counter here preemptively. */
934     if (!drop) {
935         bdrv_ref(bs);
936     }
937 
938     run_block_job(bs->job, &local_err);
939     if (local_err) {
940         goto unref_backing;
941     }
942 
943     if (!drop && bs->drv->bdrv_make_empty) {
944         ret = bs->drv->bdrv_make_empty(bs);
945         if (ret) {
946             error_setg_errno(&local_err, -ret, "Could not empty %s",
947                              filename);
948             goto unref_backing;
949         }
950     }
951 
952 unref_backing:
953     if (!drop) {
954         bdrv_unref(bs);
955     }
956 
957 done:
958     qemu_progress_end();
959 
960     blk_unref(blk);
961 
962     if (local_err) {
963         error_report_err(local_err);
964         return 1;
965     }
966 
967     qprintf(quiet, "Image committed.\n");
968     return 0;
969 }
970 
971 /*
972  * Returns true iff the first sector pointed to by 'buf' contains at least
973  * a non-NUL byte.
974  *
975  * 'pnum' is set to the number of sectors (including and immediately following
976  * the first one) that are known to be in the same allocated/unallocated state.
977  */
978 static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum)
979 {
980     bool is_zero;
981     int i;
982 
983     if (n <= 0) {
984         *pnum = 0;
985         return 0;
986     }
987     is_zero = buffer_is_zero(buf, 512);
988     for(i = 1; i < n; i++) {
989         buf += 512;
990         if (is_zero != buffer_is_zero(buf, 512)) {
991             break;
992         }
993     }
994     *pnum = i;
995     return !is_zero;
996 }
997 
998 /*
999  * Like is_allocated_sectors, but if the buffer starts with a used sector,
1000  * up to 'min' consecutive sectors containing zeros are ignored. This avoids
1001  * breaking up write requests for only small sparse areas.
1002  */
1003 static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum,
1004     int min)
1005 {
1006     int ret;
1007     int num_checked, num_used;
1008 
1009     if (n < min) {
1010         min = n;
1011     }
1012 
1013     ret = is_allocated_sectors(buf, n, pnum);
1014     if (!ret) {
1015         return ret;
1016     }
1017 
1018     num_used = *pnum;
1019     buf += BDRV_SECTOR_SIZE * *pnum;
1020     n -= *pnum;
1021     num_checked = num_used;
1022 
1023     while (n > 0) {
1024         ret = is_allocated_sectors(buf, n, pnum);
1025 
1026         buf += BDRV_SECTOR_SIZE * *pnum;
1027         n -= *pnum;
1028         num_checked += *pnum;
1029         if (ret) {
1030             num_used = num_checked;
1031         } else if (*pnum >= min) {
1032             break;
1033         }
1034     }
1035 
1036     *pnum = num_used;
1037     return 1;
1038 }
1039 
1040 /*
1041  * Compares two buffers sector by sector. Returns 0 if the first sector of both
1042  * buffers matches, non-zero otherwise.
1043  *
1044  * pnum is set to the number of sectors (including and immediately following
1045  * the first one) that are known to have the same comparison result
1046  */
1047 static int compare_sectors(const uint8_t *buf1, const uint8_t *buf2, int n,
1048     int *pnum)
1049 {
1050     bool res;
1051     int i;
1052 
1053     if (n <= 0) {
1054         *pnum = 0;
1055         return 0;
1056     }
1057 
1058     res = !!memcmp(buf1, buf2, 512);
1059     for(i = 1; i < n; i++) {
1060         buf1 += 512;
1061         buf2 += 512;
1062 
1063         if (!!memcmp(buf1, buf2, 512) != res) {
1064             break;
1065         }
1066     }
1067 
1068     *pnum = i;
1069     return res;
1070 }
1071 
1072 #define IO_BUF_SIZE (2 * 1024 * 1024)
1073 
1074 static int64_t sectors_to_bytes(int64_t sectors)
1075 {
1076     return sectors << BDRV_SECTOR_BITS;
1077 }
1078 
1079 static int64_t sectors_to_process(int64_t total, int64_t from)
1080 {
1081     return MIN(total - from, IO_BUF_SIZE >> BDRV_SECTOR_BITS);
1082 }
1083 
1084 /*
1085  * Check if passed sectors are empty (not allocated or contain only 0 bytes)
1086  *
1087  * Returns 0 in case sectors are filled with 0, 1 if sectors contain non-zero
1088  * data and negative value on error.
1089  *
1090  * @param blk:  BlockBackend for the image
1091  * @param sect_num: Number of first sector to check
1092  * @param sect_count: Number of sectors to check
1093  * @param filename: Name of disk file we are checking (logging purpose)
1094  * @param buffer: Allocated buffer for storing read data
1095  * @param quiet: Flag for quiet mode
1096  */
1097 static int check_empty_sectors(BlockBackend *blk, int64_t sect_num,
1098                                int sect_count, const char *filename,
1099                                uint8_t *buffer, bool quiet)
1100 {
1101     int pnum, ret = 0;
1102     ret = blk_pread(blk, sect_num << BDRV_SECTOR_BITS, buffer,
1103                     sect_count << BDRV_SECTOR_BITS);
1104     if (ret < 0) {
1105         error_report("Error while reading offset %" PRId64 " of %s: %s",
1106                      sectors_to_bytes(sect_num), filename, strerror(-ret));
1107         return ret;
1108     }
1109     ret = is_allocated_sectors(buffer, sect_count, &pnum);
1110     if (ret || pnum != sect_count) {
1111         qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1112                 sectors_to_bytes(ret ? sect_num : sect_num + pnum));
1113         return 1;
1114     }
1115 
1116     return 0;
1117 }
1118 
1119 /*
1120  * Compares two images. Exit codes:
1121  *
1122  * 0 - Images are identical
1123  * 1 - Images differ
1124  * >1 - Error occurred
1125  */
1126 static int img_compare(int argc, char **argv)
1127 {
1128     const char *fmt1 = NULL, *fmt2 = NULL, *cache, *filename1, *filename2;
1129     BlockBackend *blk1, *blk2;
1130     BlockDriverState *bs1, *bs2;
1131     int64_t total_sectors1, total_sectors2;
1132     uint8_t *buf1 = NULL, *buf2 = NULL;
1133     int pnum1, pnum2;
1134     int allocated1, allocated2;
1135     int ret = 0; /* return value - 0 Ident, 1 Different, >1 Error */
1136     bool progress = false, quiet = false, strict = false;
1137     int flags;
1138     bool writethrough;
1139     int64_t total_sectors;
1140     int64_t sector_num = 0;
1141     int64_t nb_sectors;
1142     int c, pnum;
1143     uint64_t progress_base;
1144     bool image_opts = false;
1145 
1146     cache = BDRV_DEFAULT_CACHE;
1147     for (;;) {
1148         static const struct option long_options[] = {
1149             {"help", no_argument, 0, 'h'},
1150             {"object", required_argument, 0, OPTION_OBJECT},
1151             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
1152             {0, 0, 0, 0}
1153         };
1154         c = getopt_long(argc, argv, "hf:F:T:pqs",
1155                         long_options, NULL);
1156         if (c == -1) {
1157             break;
1158         }
1159         switch (c) {
1160         case '?':
1161         case 'h':
1162             help();
1163             break;
1164         case 'f':
1165             fmt1 = optarg;
1166             break;
1167         case 'F':
1168             fmt2 = optarg;
1169             break;
1170         case 'T':
1171             cache = optarg;
1172             break;
1173         case 'p':
1174             progress = true;
1175             break;
1176         case 'q':
1177             quiet = true;
1178             break;
1179         case 's':
1180             strict = true;
1181             break;
1182         case OPTION_OBJECT: {
1183             QemuOpts *opts;
1184             opts = qemu_opts_parse_noisily(&qemu_object_opts,
1185                                            optarg, true);
1186             if (!opts) {
1187                 ret = 2;
1188                 goto out4;
1189             }
1190         }   break;
1191         case OPTION_IMAGE_OPTS:
1192             image_opts = true;
1193             break;
1194         }
1195     }
1196 
1197     /* Progress is not shown in Quiet mode */
1198     if (quiet) {
1199         progress = false;
1200     }
1201 
1202 
1203     if (optind != argc - 2) {
1204         error_exit("Expecting two image file names");
1205     }
1206     filename1 = argv[optind++];
1207     filename2 = argv[optind++];
1208 
1209     if (qemu_opts_foreach(&qemu_object_opts,
1210                           user_creatable_add_opts_foreach,
1211                           NULL, NULL)) {
1212         ret = 2;
1213         goto out4;
1214     }
1215 
1216     /* Initialize before goto out */
1217     qemu_progress_init(progress, 2.0);
1218 
1219     flags = 0;
1220     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
1221     if (ret < 0) {
1222         error_report("Invalid source cache option: %s", cache);
1223         ret = 2;
1224         goto out3;
1225     }
1226 
1227     blk1 = img_open(image_opts, filename1, fmt1, flags, writethrough, quiet);
1228     if (!blk1) {
1229         ret = 2;
1230         goto out3;
1231     }
1232 
1233     blk2 = img_open(image_opts, filename2, fmt2, flags, writethrough, quiet);
1234     if (!blk2) {
1235         ret = 2;
1236         goto out2;
1237     }
1238     bs1 = blk_bs(blk1);
1239     bs2 = blk_bs(blk2);
1240 
1241     buf1 = blk_blockalign(blk1, IO_BUF_SIZE);
1242     buf2 = blk_blockalign(blk2, IO_BUF_SIZE);
1243     total_sectors1 = blk_nb_sectors(blk1);
1244     if (total_sectors1 < 0) {
1245         error_report("Can't get size of %s: %s",
1246                      filename1, strerror(-total_sectors1));
1247         ret = 4;
1248         goto out;
1249     }
1250     total_sectors2 = blk_nb_sectors(blk2);
1251     if (total_sectors2 < 0) {
1252         error_report("Can't get size of %s: %s",
1253                      filename2, strerror(-total_sectors2));
1254         ret = 4;
1255         goto out;
1256     }
1257     total_sectors = MIN(total_sectors1, total_sectors2);
1258     progress_base = MAX(total_sectors1, total_sectors2);
1259 
1260     qemu_progress_print(0, 100);
1261 
1262     if (strict && total_sectors1 != total_sectors2) {
1263         ret = 1;
1264         qprintf(quiet, "Strict mode: Image size mismatch!\n");
1265         goto out;
1266     }
1267 
1268     for (;;) {
1269         int64_t status1, status2;
1270         BlockDriverState *file;
1271 
1272         nb_sectors = sectors_to_process(total_sectors, sector_num);
1273         if (nb_sectors <= 0) {
1274             break;
1275         }
1276         status1 = bdrv_get_block_status_above(bs1, NULL, sector_num,
1277                                               total_sectors1 - sector_num,
1278                                               &pnum1, &file);
1279         if (status1 < 0) {
1280             ret = 3;
1281             error_report("Sector allocation test failed for %s", filename1);
1282             goto out;
1283         }
1284         allocated1 = status1 & BDRV_BLOCK_ALLOCATED;
1285 
1286         status2 = bdrv_get_block_status_above(bs2, NULL, sector_num,
1287                                               total_sectors2 - sector_num,
1288                                               &pnum2, &file);
1289         if (status2 < 0) {
1290             ret = 3;
1291             error_report("Sector allocation test failed for %s", filename2);
1292             goto out;
1293         }
1294         allocated2 = status2 & BDRV_BLOCK_ALLOCATED;
1295         if (pnum1) {
1296             nb_sectors = MIN(nb_sectors, pnum1);
1297         }
1298         if (pnum2) {
1299             nb_sectors = MIN(nb_sectors, pnum2);
1300         }
1301 
1302         if (strict) {
1303             if ((status1 & ~BDRV_BLOCK_OFFSET_MASK) !=
1304                 (status2 & ~BDRV_BLOCK_OFFSET_MASK)) {
1305                 ret = 1;
1306                 qprintf(quiet, "Strict mode: Offset %" PRId64
1307                         " block status mismatch!\n",
1308                         sectors_to_bytes(sector_num));
1309                 goto out;
1310             }
1311         }
1312         if ((status1 & BDRV_BLOCK_ZERO) && (status2 & BDRV_BLOCK_ZERO)) {
1313             nb_sectors = MIN(pnum1, pnum2);
1314         } else if (allocated1 == allocated2) {
1315             if (allocated1) {
1316                 ret = blk_pread(blk1, sector_num << BDRV_SECTOR_BITS, buf1,
1317                                 nb_sectors << BDRV_SECTOR_BITS);
1318                 if (ret < 0) {
1319                     error_report("Error while reading offset %" PRId64 " of %s:"
1320                                  " %s", sectors_to_bytes(sector_num), filename1,
1321                                  strerror(-ret));
1322                     ret = 4;
1323                     goto out;
1324                 }
1325                 ret = blk_pread(blk2, sector_num << BDRV_SECTOR_BITS, buf2,
1326                                 nb_sectors << BDRV_SECTOR_BITS);
1327                 if (ret < 0) {
1328                     error_report("Error while reading offset %" PRId64
1329                                  " of %s: %s", sectors_to_bytes(sector_num),
1330                                  filename2, strerror(-ret));
1331                     ret = 4;
1332                     goto out;
1333                 }
1334                 ret = compare_sectors(buf1, buf2, nb_sectors, &pnum);
1335                 if (ret || pnum != nb_sectors) {
1336                     qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1337                             sectors_to_bytes(
1338                                 ret ? sector_num : sector_num + pnum));
1339                     ret = 1;
1340                     goto out;
1341                 }
1342             }
1343         } else {
1344 
1345             if (allocated1) {
1346                 ret = check_empty_sectors(blk1, sector_num, nb_sectors,
1347                                           filename1, buf1, quiet);
1348             } else {
1349                 ret = check_empty_sectors(blk2, sector_num, nb_sectors,
1350                                           filename2, buf1, quiet);
1351             }
1352             if (ret) {
1353                 if (ret < 0) {
1354                     error_report("Error while reading offset %" PRId64 ": %s",
1355                                  sectors_to_bytes(sector_num), strerror(-ret));
1356                     ret = 4;
1357                 }
1358                 goto out;
1359             }
1360         }
1361         sector_num += nb_sectors;
1362         qemu_progress_print(((float) nb_sectors / progress_base)*100, 100);
1363     }
1364 
1365     if (total_sectors1 != total_sectors2) {
1366         BlockBackend *blk_over;
1367         int64_t total_sectors_over;
1368         const char *filename_over;
1369 
1370         qprintf(quiet, "Warning: Image size mismatch!\n");
1371         if (total_sectors1 > total_sectors2) {
1372             total_sectors_over = total_sectors1;
1373             blk_over = blk1;
1374             filename_over = filename1;
1375         } else {
1376             total_sectors_over = total_sectors2;
1377             blk_over = blk2;
1378             filename_over = filename2;
1379         }
1380 
1381         for (;;) {
1382             nb_sectors = sectors_to_process(total_sectors_over, sector_num);
1383             if (nb_sectors <= 0) {
1384                 break;
1385             }
1386             ret = bdrv_is_allocated_above(blk_bs(blk_over), NULL, sector_num,
1387                                           nb_sectors, &pnum);
1388             if (ret < 0) {
1389                 ret = 3;
1390                 error_report("Sector allocation test failed for %s",
1391                              filename_over);
1392                 goto out;
1393 
1394             }
1395             nb_sectors = pnum;
1396             if (ret) {
1397                 ret = check_empty_sectors(blk_over, sector_num, nb_sectors,
1398                                           filename_over, buf1, quiet);
1399                 if (ret) {
1400                     if (ret < 0) {
1401                         error_report("Error while reading offset %" PRId64
1402                                      " of %s: %s", sectors_to_bytes(sector_num),
1403                                      filename_over, strerror(-ret));
1404                         ret = 4;
1405                     }
1406                     goto out;
1407                 }
1408             }
1409             sector_num += nb_sectors;
1410             qemu_progress_print(((float) nb_sectors / progress_base)*100, 100);
1411         }
1412     }
1413 
1414     qprintf(quiet, "Images are identical.\n");
1415     ret = 0;
1416 
1417 out:
1418     qemu_vfree(buf1);
1419     qemu_vfree(buf2);
1420     blk_unref(blk2);
1421 out2:
1422     blk_unref(blk1);
1423 out3:
1424     qemu_progress_end();
1425 out4:
1426     return ret;
1427 }
1428 
1429 enum ImgConvertBlockStatus {
1430     BLK_DATA,
1431     BLK_ZERO,
1432     BLK_BACKING_FILE,
1433 };
1434 
1435 typedef struct ImgConvertState {
1436     BlockBackend **src;
1437     int64_t *src_sectors;
1438     int src_cur, src_num;
1439     int64_t src_cur_offset;
1440     int64_t total_sectors;
1441     int64_t allocated_sectors;
1442     enum ImgConvertBlockStatus status;
1443     int64_t sector_next_status;
1444     BlockBackend *target;
1445     bool has_zero_init;
1446     bool compressed;
1447     bool target_has_backing;
1448     int min_sparse;
1449     size_t cluster_sectors;
1450     size_t buf_sectors;
1451 } ImgConvertState;
1452 
1453 static void convert_select_part(ImgConvertState *s, int64_t sector_num)
1454 {
1455     assert(sector_num >= s->src_cur_offset);
1456     while (sector_num - s->src_cur_offset >= s->src_sectors[s->src_cur]) {
1457         s->src_cur_offset += s->src_sectors[s->src_cur];
1458         s->src_cur++;
1459         assert(s->src_cur < s->src_num);
1460     }
1461 }
1462 
1463 static int convert_iteration_sectors(ImgConvertState *s, int64_t sector_num)
1464 {
1465     int64_t ret;
1466     int n;
1467 
1468     convert_select_part(s, sector_num);
1469 
1470     assert(s->total_sectors > sector_num);
1471     n = MIN(s->total_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
1472 
1473     if (s->sector_next_status <= sector_num) {
1474         BlockDriverState *file;
1475         ret = bdrv_get_block_status(blk_bs(s->src[s->src_cur]),
1476                                     sector_num - s->src_cur_offset,
1477                                     n, &n, &file);
1478         if (ret < 0) {
1479             return ret;
1480         }
1481 
1482         if (ret & BDRV_BLOCK_ZERO) {
1483             s->status = BLK_ZERO;
1484         } else if (ret & BDRV_BLOCK_DATA) {
1485             s->status = BLK_DATA;
1486         } else if (!s->target_has_backing) {
1487             /* Without a target backing file we must copy over the contents of
1488              * the backing file as well. */
1489             /* Check block status of the backing file chain to avoid
1490              * needlessly reading zeroes and limiting the iteration to the
1491              * buffer size */
1492             ret = bdrv_get_block_status_above(blk_bs(s->src[s->src_cur]), NULL,
1493                                               sector_num - s->src_cur_offset,
1494                                               n, &n, &file);
1495             if (ret < 0) {
1496                 return ret;
1497             }
1498 
1499             if (ret & BDRV_BLOCK_ZERO) {
1500                 s->status = BLK_ZERO;
1501             } else {
1502                 s->status = BLK_DATA;
1503             }
1504         } else {
1505             s->status = BLK_BACKING_FILE;
1506         }
1507 
1508         s->sector_next_status = sector_num + n;
1509     }
1510 
1511     n = MIN(n, s->sector_next_status - sector_num);
1512     if (s->status == BLK_DATA) {
1513         n = MIN(n, s->buf_sectors);
1514     }
1515 
1516     /* We need to write complete clusters for compressed images, so if an
1517      * unallocated area is shorter than that, we must consider the whole
1518      * cluster allocated. */
1519     if (s->compressed) {
1520         if (n < s->cluster_sectors) {
1521             n = MIN(s->cluster_sectors, s->total_sectors - sector_num);
1522             s->status = BLK_DATA;
1523         } else {
1524             n = QEMU_ALIGN_DOWN(n, s->cluster_sectors);
1525         }
1526     }
1527 
1528     return n;
1529 }
1530 
1531 static int convert_read(ImgConvertState *s, int64_t sector_num, int nb_sectors,
1532                         uint8_t *buf)
1533 {
1534     int n;
1535     int ret;
1536 
1537     assert(nb_sectors <= s->buf_sectors);
1538     while (nb_sectors > 0) {
1539         BlockBackend *blk;
1540         int64_t bs_sectors;
1541 
1542         /* In the case of compression with multiple source files, we can get a
1543          * nb_sectors that spreads into the next part. So we must be able to
1544          * read across multiple BDSes for one convert_read() call. */
1545         convert_select_part(s, sector_num);
1546         blk = s->src[s->src_cur];
1547         bs_sectors = s->src_sectors[s->src_cur];
1548 
1549         n = MIN(nb_sectors, bs_sectors - (sector_num - s->src_cur_offset));
1550         ret = blk_pread(blk,
1551                         (sector_num - s->src_cur_offset) << BDRV_SECTOR_BITS,
1552                         buf, n << BDRV_SECTOR_BITS);
1553         if (ret < 0) {
1554             return ret;
1555         }
1556 
1557         sector_num += n;
1558         nb_sectors -= n;
1559         buf += n * BDRV_SECTOR_SIZE;
1560     }
1561 
1562     return 0;
1563 }
1564 
1565 static int convert_write(ImgConvertState *s, int64_t sector_num, int nb_sectors,
1566                          const uint8_t *buf)
1567 {
1568     int ret;
1569 
1570     while (nb_sectors > 0) {
1571         int n = nb_sectors;
1572 
1573         switch (s->status) {
1574         case BLK_BACKING_FILE:
1575             /* If we have a backing file, leave clusters unallocated that are
1576              * unallocated in the source image, so that the backing file is
1577              * visible at the respective offset. */
1578             assert(s->target_has_backing);
1579             break;
1580 
1581         case BLK_DATA:
1582             /* We must always write compressed clusters as a whole, so don't
1583              * try to find zeroed parts in the buffer. We can only save the
1584              * write if the buffer is completely zeroed and we're allowed to
1585              * keep the target sparse. */
1586             if (s->compressed) {
1587                 if (s->has_zero_init && s->min_sparse &&
1588                     buffer_is_zero(buf, n * BDRV_SECTOR_SIZE))
1589                 {
1590                     assert(!s->target_has_backing);
1591                     break;
1592                 }
1593 
1594                 ret = blk_write_compressed(s->target, sector_num, buf, n);
1595                 if (ret < 0) {
1596                     return ret;
1597                 }
1598                 break;
1599             }
1600 
1601             /* If there is real non-zero data or we're told to keep the target
1602              * fully allocated (-S 0), we must write it. Otherwise we can treat
1603              * it as zero sectors. */
1604             if (!s->min_sparse ||
1605                 is_allocated_sectors_min(buf, n, &n, s->min_sparse))
1606             {
1607                 ret = blk_pwrite(s->target, sector_num << BDRV_SECTOR_BITS,
1608                                  buf, n << BDRV_SECTOR_BITS, 0);
1609                 if (ret < 0) {
1610                     return ret;
1611                 }
1612                 break;
1613             }
1614             /* fall-through */
1615 
1616         case BLK_ZERO:
1617             if (s->has_zero_init) {
1618                 break;
1619             }
1620             ret = blk_pwrite_zeroes(s->target, sector_num << BDRV_SECTOR_BITS,
1621                                     n << BDRV_SECTOR_BITS, 0);
1622             if (ret < 0) {
1623                 return ret;
1624             }
1625             break;
1626         }
1627 
1628         sector_num += n;
1629         nb_sectors -= n;
1630         buf += n * BDRV_SECTOR_SIZE;
1631     }
1632 
1633     return 0;
1634 }
1635 
1636 static int convert_do_copy(ImgConvertState *s)
1637 {
1638     uint8_t *buf = NULL;
1639     int64_t sector_num, allocated_done;
1640     int ret;
1641     int n;
1642 
1643     /* Check whether we have zero initialisation or can get it efficiently */
1644     s->has_zero_init = s->min_sparse && !s->target_has_backing
1645                      ? bdrv_has_zero_init(blk_bs(s->target))
1646                      : false;
1647 
1648     if (!s->has_zero_init && !s->target_has_backing &&
1649         bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)))
1650     {
1651         ret = bdrv_make_zero(blk_bs(s->target), BDRV_REQ_MAY_UNMAP);
1652         if (ret == 0) {
1653             s->has_zero_init = true;
1654         }
1655     }
1656 
1657     /* Allocate buffer for copied data. For compressed images, only one cluster
1658      * can be copied at a time. */
1659     if (s->compressed) {
1660         if (s->cluster_sectors <= 0 || s->cluster_sectors > s->buf_sectors) {
1661             error_report("invalid cluster size");
1662             ret = -EINVAL;
1663             goto fail;
1664         }
1665         s->buf_sectors = s->cluster_sectors;
1666     }
1667     buf = blk_blockalign(s->target, s->buf_sectors * BDRV_SECTOR_SIZE);
1668 
1669     /* Calculate allocated sectors for progress */
1670     s->allocated_sectors = 0;
1671     sector_num = 0;
1672     while (sector_num < s->total_sectors) {
1673         n = convert_iteration_sectors(s, sector_num);
1674         if (n < 0) {
1675             ret = n;
1676             goto fail;
1677         }
1678         if (s->status == BLK_DATA || (!s->min_sparse && s->status == BLK_ZERO))
1679         {
1680             s->allocated_sectors += n;
1681         }
1682         sector_num += n;
1683     }
1684 
1685     /* Do the copy */
1686     s->src_cur = 0;
1687     s->src_cur_offset = 0;
1688     s->sector_next_status = 0;
1689 
1690     sector_num = 0;
1691     allocated_done = 0;
1692 
1693     while (sector_num < s->total_sectors) {
1694         n = convert_iteration_sectors(s, sector_num);
1695         if (n < 0) {
1696             ret = n;
1697             goto fail;
1698         }
1699         if (s->status == BLK_DATA || (!s->min_sparse && s->status == BLK_ZERO))
1700         {
1701             allocated_done += n;
1702             qemu_progress_print(100.0 * allocated_done / s->allocated_sectors,
1703                                 0);
1704         }
1705 
1706         if (s->status == BLK_DATA) {
1707             ret = convert_read(s, sector_num, n, buf);
1708             if (ret < 0) {
1709                 error_report("error while reading sector %" PRId64
1710                              ": %s", sector_num, strerror(-ret));
1711                 goto fail;
1712             }
1713         } else if (!s->min_sparse && s->status == BLK_ZERO) {
1714             n = MIN(n, s->buf_sectors);
1715             memset(buf, 0, n * BDRV_SECTOR_SIZE);
1716             s->status = BLK_DATA;
1717         }
1718 
1719         ret = convert_write(s, sector_num, n, buf);
1720         if (ret < 0) {
1721             error_report("error while writing sector %" PRId64
1722                          ": %s", sector_num, strerror(-ret));
1723             goto fail;
1724         }
1725 
1726         sector_num += n;
1727     }
1728 
1729     if (s->compressed) {
1730         /* signal EOF to align */
1731         ret = blk_write_compressed(s->target, 0, NULL, 0);
1732         if (ret < 0) {
1733             goto fail;
1734         }
1735     }
1736 
1737     ret = 0;
1738 fail:
1739     qemu_vfree(buf);
1740     return ret;
1741 }
1742 
1743 static int img_convert(int argc, char **argv)
1744 {
1745     int c, bs_n, bs_i, compress, cluster_sectors, skip_create;
1746     int64_t ret = 0;
1747     int progress = 0, flags, src_flags;
1748     bool writethrough, src_writethrough;
1749     const char *fmt, *out_fmt, *cache, *src_cache, *out_baseimg, *out_filename;
1750     BlockDriver *drv, *proto_drv;
1751     BlockBackend **blk = NULL, *out_blk = NULL;
1752     BlockDriverState **bs = NULL, *out_bs = NULL;
1753     int64_t total_sectors;
1754     int64_t *bs_sectors = NULL;
1755     size_t bufsectors = IO_BUF_SIZE / BDRV_SECTOR_SIZE;
1756     BlockDriverInfo bdi;
1757     QemuOpts *opts = NULL;
1758     QemuOptsList *create_opts = NULL;
1759     const char *out_baseimg_param;
1760     char *options = NULL;
1761     const char *snapshot_name = NULL;
1762     int min_sparse = 8; /* Need at least 4k of zeros for sparse detection */
1763     bool quiet = false;
1764     Error *local_err = NULL;
1765     QemuOpts *sn_opts = NULL;
1766     ImgConvertState state;
1767     bool image_opts = false;
1768 
1769     fmt = NULL;
1770     out_fmt = "raw";
1771     cache = "unsafe";
1772     src_cache = BDRV_DEFAULT_CACHE;
1773     out_baseimg = NULL;
1774     compress = 0;
1775     skip_create = 0;
1776     for(;;) {
1777         static const struct option long_options[] = {
1778             {"help", no_argument, 0, 'h'},
1779             {"object", required_argument, 0, OPTION_OBJECT},
1780             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
1781             {0, 0, 0, 0}
1782         };
1783         c = getopt_long(argc, argv, "hf:O:B:ce6o:s:l:S:pt:T:qn",
1784                         long_options, NULL);
1785         if (c == -1) {
1786             break;
1787         }
1788         switch(c) {
1789         case '?':
1790         case 'h':
1791             help();
1792             break;
1793         case 'f':
1794             fmt = optarg;
1795             break;
1796         case 'O':
1797             out_fmt = optarg;
1798             break;
1799         case 'B':
1800             out_baseimg = optarg;
1801             break;
1802         case 'c':
1803             compress = 1;
1804             break;
1805         case 'e':
1806             error_report("option -e is deprecated, please use \'-o "
1807                   "encryption\' instead!");
1808             ret = -1;
1809             goto fail_getopt;
1810         case '6':
1811             error_report("option -6 is deprecated, please use \'-o "
1812                   "compat6\' instead!");
1813             ret = -1;
1814             goto fail_getopt;
1815         case 'o':
1816             if (!is_valid_option_list(optarg)) {
1817                 error_report("Invalid option list: %s", optarg);
1818                 ret = -1;
1819                 goto fail_getopt;
1820             }
1821             if (!options) {
1822                 options = g_strdup(optarg);
1823             } else {
1824                 char *old_options = options;
1825                 options = g_strdup_printf("%s,%s", options, optarg);
1826                 g_free(old_options);
1827             }
1828             break;
1829         case 's':
1830             snapshot_name = optarg;
1831             break;
1832         case 'l':
1833             if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
1834                 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
1835                                                   optarg, false);
1836                 if (!sn_opts) {
1837                     error_report("Failed in parsing snapshot param '%s'",
1838                                  optarg);
1839                     ret = -1;
1840                     goto fail_getopt;
1841                 }
1842             } else {
1843                 snapshot_name = optarg;
1844             }
1845             break;
1846         case 'S':
1847         {
1848             int64_t sval;
1849             char *end;
1850             sval = qemu_strtosz_suffix(optarg, &end, QEMU_STRTOSZ_DEFSUFFIX_B);
1851             if (sval < 0 || *end) {
1852                 error_report("Invalid minimum zero buffer size for sparse output specified");
1853                 ret = -1;
1854                 goto fail_getopt;
1855             }
1856 
1857             min_sparse = sval / BDRV_SECTOR_SIZE;
1858             break;
1859         }
1860         case 'p':
1861             progress = 1;
1862             break;
1863         case 't':
1864             cache = optarg;
1865             break;
1866         case 'T':
1867             src_cache = optarg;
1868             break;
1869         case 'q':
1870             quiet = true;
1871             break;
1872         case 'n':
1873             skip_create = 1;
1874             break;
1875         case OPTION_OBJECT:
1876             opts = qemu_opts_parse_noisily(&qemu_object_opts,
1877                                            optarg, true);
1878             if (!opts) {
1879                 goto fail_getopt;
1880             }
1881             break;
1882         case OPTION_IMAGE_OPTS:
1883             image_opts = true;
1884             break;
1885         }
1886     }
1887 
1888     if (qemu_opts_foreach(&qemu_object_opts,
1889                           user_creatable_add_opts_foreach,
1890                           NULL, NULL)) {
1891         goto fail_getopt;
1892     }
1893 
1894     /* Initialize before goto out */
1895     if (quiet) {
1896         progress = 0;
1897     }
1898     qemu_progress_init(progress, 1.0);
1899 
1900     bs_n = argc - optind - 1;
1901     out_filename = bs_n >= 1 ? argv[argc - 1] : NULL;
1902 
1903     if (options && has_help_option(options)) {
1904         ret = print_block_option_help(out_filename, out_fmt);
1905         goto out;
1906     }
1907 
1908     if (bs_n < 1) {
1909         error_exit("Must specify image file name");
1910     }
1911 
1912 
1913     if (bs_n > 1 && out_baseimg) {
1914         error_report("-B makes no sense when concatenating multiple input "
1915                      "images");
1916         ret = -1;
1917         goto out;
1918     }
1919 
1920     src_flags = 0;
1921     ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
1922     if (ret < 0) {
1923         error_report("Invalid source cache option: %s", src_cache);
1924         goto out;
1925     }
1926 
1927     qemu_progress_print(0, 100);
1928 
1929     blk = g_new0(BlockBackend *, bs_n);
1930     bs = g_new0(BlockDriverState *, bs_n);
1931     bs_sectors = g_new(int64_t, bs_n);
1932 
1933     total_sectors = 0;
1934     for (bs_i = 0; bs_i < bs_n; bs_i++) {
1935         blk[bs_i] = img_open(image_opts, argv[optind + bs_i],
1936                              fmt, src_flags, src_writethrough, quiet);
1937         if (!blk[bs_i]) {
1938             ret = -1;
1939             goto out;
1940         }
1941         bs[bs_i] = blk_bs(blk[bs_i]);
1942         bs_sectors[bs_i] = blk_nb_sectors(blk[bs_i]);
1943         if (bs_sectors[bs_i] < 0) {
1944             error_report("Could not get size of %s: %s",
1945                          argv[optind + bs_i], strerror(-bs_sectors[bs_i]));
1946             ret = -1;
1947             goto out;
1948         }
1949         total_sectors += bs_sectors[bs_i];
1950     }
1951 
1952     if (sn_opts) {
1953         ret = bdrv_snapshot_load_tmp(bs[0],
1954                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
1955                                      qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
1956                                      &local_err);
1957     } else if (snapshot_name != NULL) {
1958         if (bs_n > 1) {
1959             error_report("No support for concatenating multiple snapshot");
1960             ret = -1;
1961             goto out;
1962         }
1963 
1964         bdrv_snapshot_load_tmp_by_id_or_name(bs[0], snapshot_name, &local_err);
1965     }
1966     if (local_err) {
1967         error_reportf_err(local_err, "Failed to load snapshot: ");
1968         ret = -1;
1969         goto out;
1970     }
1971 
1972     /* Find driver and parse its options */
1973     drv = bdrv_find_format(out_fmt);
1974     if (!drv) {
1975         error_report("Unknown file format '%s'", out_fmt);
1976         ret = -1;
1977         goto out;
1978     }
1979 
1980     proto_drv = bdrv_find_protocol(out_filename, true, &local_err);
1981     if (!proto_drv) {
1982         error_report_err(local_err);
1983         ret = -1;
1984         goto out;
1985     }
1986 
1987     if (!skip_create) {
1988         if (!drv->create_opts) {
1989             error_report("Format driver '%s' does not support image creation",
1990                          drv->format_name);
1991             ret = -1;
1992             goto out;
1993         }
1994 
1995         if (!proto_drv->create_opts) {
1996             error_report("Protocol driver '%s' does not support image creation",
1997                          proto_drv->format_name);
1998             ret = -1;
1999             goto out;
2000         }
2001 
2002         create_opts = qemu_opts_append(create_opts, drv->create_opts);
2003         create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
2004 
2005         opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
2006         if (options) {
2007             qemu_opts_do_parse(opts, options, NULL, &local_err);
2008             if (local_err) {
2009                 error_report_err(local_err);
2010                 ret = -1;
2011                 goto out;
2012             }
2013         }
2014 
2015         qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_sectors * 512,
2016                             &error_abort);
2017         ret = add_old_style_options(out_fmt, opts, out_baseimg, NULL);
2018         if (ret < 0) {
2019             goto out;
2020         }
2021     }
2022 
2023     /* Get backing file name if -o backing_file was used */
2024     out_baseimg_param = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
2025     if (out_baseimg_param) {
2026         out_baseimg = out_baseimg_param;
2027     }
2028 
2029     /* Check if compression is supported */
2030     if (compress) {
2031         bool encryption =
2032             qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, false);
2033         const char *preallocation =
2034             qemu_opt_get(opts, BLOCK_OPT_PREALLOC);
2035 
2036         if (!drv->bdrv_write_compressed) {
2037             error_report("Compression not supported for this file format");
2038             ret = -1;
2039             goto out;
2040         }
2041 
2042         if (encryption) {
2043             error_report("Compression and encryption not supported at "
2044                          "the same time");
2045             ret = -1;
2046             goto out;
2047         }
2048 
2049         if (preallocation
2050             && strcmp(preallocation, "off"))
2051         {
2052             error_report("Compression and preallocation not supported at "
2053                          "the same time");
2054             ret = -1;
2055             goto out;
2056         }
2057     }
2058 
2059     if (!skip_create) {
2060         /* Create the new image */
2061         ret = bdrv_create(drv, out_filename, opts, &local_err);
2062         if (ret < 0) {
2063             error_reportf_err(local_err, "%s: error while converting %s: ",
2064                               out_filename, out_fmt);
2065             goto out;
2066         }
2067     }
2068 
2069     flags = min_sparse ? (BDRV_O_RDWR | BDRV_O_UNMAP) : BDRV_O_RDWR;
2070     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
2071     if (ret < 0) {
2072         error_report("Invalid cache option: %s", cache);
2073         goto out;
2074     }
2075 
2076     /* XXX we should allow --image-opts to trigger use of
2077      * img_open() here, but then we have trouble with
2078      * the bdrv_create() call which takes different params.
2079      * Not critical right now, so fix can wait...
2080      */
2081     out_blk = img_open_file(out_filename, out_fmt, flags, writethrough, quiet);
2082     if (!out_blk) {
2083         ret = -1;
2084         goto out;
2085     }
2086     out_bs = blk_bs(out_blk);
2087 
2088     /* increase bufsectors from the default 4096 (2M) if opt_transfer_length
2089      * or discard_alignment of the out_bs is greater. Limit to 32768 (16MB)
2090      * as maximum. */
2091     bufsectors = MIN(32768,
2092                      MAX(bufsectors, MAX(out_bs->bl.opt_transfer_length,
2093                                          out_bs->bl.discard_alignment))
2094                     );
2095 
2096     if (skip_create) {
2097         int64_t output_sectors = blk_nb_sectors(out_blk);
2098         if (output_sectors < 0) {
2099             error_report("unable to get output image length: %s",
2100                          strerror(-output_sectors));
2101             ret = -1;
2102             goto out;
2103         } else if (output_sectors < total_sectors) {
2104             error_report("output file is smaller than input file");
2105             ret = -1;
2106             goto out;
2107         }
2108     }
2109 
2110     cluster_sectors = 0;
2111     ret = bdrv_get_info(out_bs, &bdi);
2112     if (ret < 0) {
2113         if (compress) {
2114             error_report("could not get block driver info");
2115             goto out;
2116         }
2117     } else {
2118         compress = compress || bdi.needs_compressed_writes;
2119         cluster_sectors = bdi.cluster_size / BDRV_SECTOR_SIZE;
2120     }
2121 
2122     state = (ImgConvertState) {
2123         .src                = blk,
2124         .src_sectors        = bs_sectors,
2125         .src_num            = bs_n,
2126         .total_sectors      = total_sectors,
2127         .target             = out_blk,
2128         .compressed         = compress,
2129         .target_has_backing = (bool) out_baseimg,
2130         .min_sparse         = min_sparse,
2131         .cluster_sectors    = cluster_sectors,
2132         .buf_sectors        = bufsectors,
2133     };
2134     ret = convert_do_copy(&state);
2135 
2136 out:
2137     if (!ret) {
2138         qemu_progress_print(100, 0);
2139     }
2140     qemu_progress_end();
2141     qemu_opts_del(opts);
2142     qemu_opts_free(create_opts);
2143     qemu_opts_del(sn_opts);
2144     blk_unref(out_blk);
2145     g_free(bs);
2146     if (blk) {
2147         for (bs_i = 0; bs_i < bs_n; bs_i++) {
2148             blk_unref(blk[bs_i]);
2149         }
2150         g_free(blk);
2151     }
2152     g_free(bs_sectors);
2153 fail_getopt:
2154     g_free(options);
2155 
2156     if (ret) {
2157         return 1;
2158     }
2159     return 0;
2160 }
2161 
2162 
2163 static void dump_snapshots(BlockDriverState *bs)
2164 {
2165     QEMUSnapshotInfo *sn_tab, *sn;
2166     int nb_sns, i;
2167 
2168     nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2169     if (nb_sns <= 0)
2170         return;
2171     printf("Snapshot list:\n");
2172     bdrv_snapshot_dump(fprintf, stdout, NULL);
2173     printf("\n");
2174     for(i = 0; i < nb_sns; i++) {
2175         sn = &sn_tab[i];
2176         bdrv_snapshot_dump(fprintf, stdout, sn);
2177         printf("\n");
2178     }
2179     g_free(sn_tab);
2180 }
2181 
2182 static void dump_json_image_info_list(ImageInfoList *list)
2183 {
2184     Error *local_err = NULL;
2185     QString *str;
2186     QmpOutputVisitor *ov = qmp_output_visitor_new();
2187     QObject *obj;
2188     visit_type_ImageInfoList(qmp_output_get_visitor(ov), NULL, &list,
2189                              &local_err);
2190     obj = qmp_output_get_qobject(ov);
2191     str = qobject_to_json_pretty(obj);
2192     assert(str != NULL);
2193     printf("%s\n", qstring_get_str(str));
2194     qobject_decref(obj);
2195     qmp_output_visitor_cleanup(ov);
2196     QDECREF(str);
2197 }
2198 
2199 static void dump_json_image_info(ImageInfo *info)
2200 {
2201     Error *local_err = NULL;
2202     QString *str;
2203     QmpOutputVisitor *ov = qmp_output_visitor_new();
2204     QObject *obj;
2205     visit_type_ImageInfo(qmp_output_get_visitor(ov), NULL, &info, &local_err);
2206     obj = qmp_output_get_qobject(ov);
2207     str = qobject_to_json_pretty(obj);
2208     assert(str != NULL);
2209     printf("%s\n", qstring_get_str(str));
2210     qobject_decref(obj);
2211     qmp_output_visitor_cleanup(ov);
2212     QDECREF(str);
2213 }
2214 
2215 static void dump_human_image_info_list(ImageInfoList *list)
2216 {
2217     ImageInfoList *elem;
2218     bool delim = false;
2219 
2220     for (elem = list; elem; elem = elem->next) {
2221         if (delim) {
2222             printf("\n");
2223         }
2224         delim = true;
2225 
2226         bdrv_image_info_dump(fprintf, stdout, elem->value);
2227     }
2228 }
2229 
2230 static gboolean str_equal_func(gconstpointer a, gconstpointer b)
2231 {
2232     return strcmp(a, b) == 0;
2233 }
2234 
2235 /**
2236  * Open an image file chain and return an ImageInfoList
2237  *
2238  * @filename: topmost image filename
2239  * @fmt: topmost image format (may be NULL to autodetect)
2240  * @chain: true  - enumerate entire backing file chain
2241  *         false - only topmost image file
2242  *
2243  * Returns a list of ImageInfo objects or NULL if there was an error opening an
2244  * image file.  If there was an error a message will have been printed to
2245  * stderr.
2246  */
2247 static ImageInfoList *collect_image_info_list(bool image_opts,
2248                                               const char *filename,
2249                                               const char *fmt,
2250                                               bool chain)
2251 {
2252     ImageInfoList *head = NULL;
2253     ImageInfoList **last = &head;
2254     GHashTable *filenames;
2255     Error *err = NULL;
2256 
2257     filenames = g_hash_table_new_full(g_str_hash, str_equal_func, NULL, NULL);
2258 
2259     while (filename) {
2260         BlockBackend *blk;
2261         BlockDriverState *bs;
2262         ImageInfo *info;
2263         ImageInfoList *elem;
2264 
2265         if (g_hash_table_lookup_extended(filenames, filename, NULL, NULL)) {
2266             error_report("Backing file '%s' creates an infinite loop.",
2267                          filename);
2268             goto err;
2269         }
2270         g_hash_table_insert(filenames, (gpointer)filename, NULL);
2271 
2272         blk = img_open(image_opts, filename, fmt,
2273                        BDRV_O_NO_BACKING | BDRV_O_NO_IO, false, false);
2274         if (!blk) {
2275             goto err;
2276         }
2277         bs = blk_bs(blk);
2278 
2279         bdrv_query_image_info(bs, &info, &err);
2280         if (err) {
2281             error_report_err(err);
2282             blk_unref(blk);
2283             goto err;
2284         }
2285 
2286         elem = g_new0(ImageInfoList, 1);
2287         elem->value = info;
2288         *last = elem;
2289         last = &elem->next;
2290 
2291         blk_unref(blk);
2292 
2293         filename = fmt = NULL;
2294         if (chain) {
2295             if (info->has_full_backing_filename) {
2296                 filename = info->full_backing_filename;
2297             } else if (info->has_backing_filename) {
2298                 error_report("Could not determine absolute backing filename,"
2299                              " but backing filename '%s' present",
2300                              info->backing_filename);
2301                 goto err;
2302             }
2303             if (info->has_backing_filename_format) {
2304                 fmt = info->backing_filename_format;
2305             }
2306         }
2307     }
2308     g_hash_table_destroy(filenames);
2309     return head;
2310 
2311 err:
2312     qapi_free_ImageInfoList(head);
2313     g_hash_table_destroy(filenames);
2314     return NULL;
2315 }
2316 
2317 static int img_info(int argc, char **argv)
2318 {
2319     int c;
2320     OutputFormat output_format = OFORMAT_HUMAN;
2321     bool chain = false;
2322     const char *filename, *fmt, *output;
2323     ImageInfoList *list;
2324     bool image_opts = false;
2325 
2326     fmt = NULL;
2327     output = NULL;
2328     for(;;) {
2329         int option_index = 0;
2330         static const struct option long_options[] = {
2331             {"help", no_argument, 0, 'h'},
2332             {"format", required_argument, 0, 'f'},
2333             {"output", required_argument, 0, OPTION_OUTPUT},
2334             {"backing-chain", no_argument, 0, OPTION_BACKING_CHAIN},
2335             {"object", required_argument, 0, OPTION_OBJECT},
2336             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2337             {0, 0, 0, 0}
2338         };
2339         c = getopt_long(argc, argv, "f:h",
2340                         long_options, &option_index);
2341         if (c == -1) {
2342             break;
2343         }
2344         switch(c) {
2345         case '?':
2346         case 'h':
2347             help();
2348             break;
2349         case 'f':
2350             fmt = optarg;
2351             break;
2352         case OPTION_OUTPUT:
2353             output = optarg;
2354             break;
2355         case OPTION_BACKING_CHAIN:
2356             chain = true;
2357             break;
2358         case OPTION_OBJECT: {
2359             QemuOpts *opts;
2360             opts = qemu_opts_parse_noisily(&qemu_object_opts,
2361                                            optarg, true);
2362             if (!opts) {
2363                 return 1;
2364             }
2365         }   break;
2366         case OPTION_IMAGE_OPTS:
2367             image_opts = true;
2368             break;
2369         }
2370     }
2371     if (optind != argc - 1) {
2372         error_exit("Expecting one image file name");
2373     }
2374     filename = argv[optind++];
2375 
2376     if (output && !strcmp(output, "json")) {
2377         output_format = OFORMAT_JSON;
2378     } else if (output && !strcmp(output, "human")) {
2379         output_format = OFORMAT_HUMAN;
2380     } else if (output) {
2381         error_report("--output must be used with human or json as argument.");
2382         return 1;
2383     }
2384 
2385     if (qemu_opts_foreach(&qemu_object_opts,
2386                           user_creatable_add_opts_foreach,
2387                           NULL, NULL)) {
2388         return 1;
2389     }
2390 
2391     list = collect_image_info_list(image_opts, filename, fmt, chain);
2392     if (!list) {
2393         return 1;
2394     }
2395 
2396     switch (output_format) {
2397     case OFORMAT_HUMAN:
2398         dump_human_image_info_list(list);
2399         break;
2400     case OFORMAT_JSON:
2401         if (chain) {
2402             dump_json_image_info_list(list);
2403         } else {
2404             dump_json_image_info(list->value);
2405         }
2406         break;
2407     }
2408 
2409     qapi_free_ImageInfoList(list);
2410     return 0;
2411 }
2412 
2413 static void dump_map_entry(OutputFormat output_format, MapEntry *e,
2414                            MapEntry *next)
2415 {
2416     switch (output_format) {
2417     case OFORMAT_HUMAN:
2418         if (e->data && !e->has_offset) {
2419             error_report("File contains external, encrypted or compressed clusters.");
2420             exit(1);
2421         }
2422         if (e->data && !e->zero) {
2423             printf("%#-16"PRIx64"%#-16"PRIx64"%#-16"PRIx64"%s\n",
2424                    e->start, e->length,
2425                    e->has_offset ? e->offset : 0,
2426                    e->has_filename ? e->filename : "");
2427         }
2428         /* This format ignores the distinction between 0, ZERO and ZERO|DATA.
2429          * Modify the flags here to allow more coalescing.
2430          */
2431         if (next && (!next->data || next->zero)) {
2432             next->data = false;
2433             next->zero = true;
2434         }
2435         break;
2436     case OFORMAT_JSON:
2437         printf("%s{ \"start\": %"PRId64", \"length\": %"PRId64","
2438                " \"depth\": %"PRId64", \"zero\": %s, \"data\": %s",
2439                (e->start == 0 ? "[" : ",\n"),
2440                e->start, e->length, e->depth,
2441                e->zero ? "true" : "false",
2442                e->data ? "true" : "false");
2443         if (e->has_offset) {
2444             printf(", \"offset\": %"PRId64"", e->offset);
2445         }
2446         putchar('}');
2447 
2448         if (!next) {
2449             printf("]\n");
2450         }
2451         break;
2452     }
2453 }
2454 
2455 static int get_block_status(BlockDriverState *bs, int64_t sector_num,
2456                             int nb_sectors, MapEntry *e)
2457 {
2458     int64_t ret;
2459     int depth;
2460     BlockDriverState *file;
2461     bool has_offset;
2462 
2463     /* As an optimization, we could cache the current range of unallocated
2464      * clusters in each file of the chain, and avoid querying the same
2465      * range repeatedly.
2466      */
2467 
2468     depth = 0;
2469     for (;;) {
2470         ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &nb_sectors,
2471                                     &file);
2472         if (ret < 0) {
2473             return ret;
2474         }
2475         assert(nb_sectors);
2476         if (ret & (BDRV_BLOCK_ZERO|BDRV_BLOCK_DATA)) {
2477             break;
2478         }
2479         bs = backing_bs(bs);
2480         if (bs == NULL) {
2481             ret = 0;
2482             break;
2483         }
2484 
2485         depth++;
2486     }
2487 
2488     has_offset = !!(ret & BDRV_BLOCK_OFFSET_VALID);
2489 
2490     *e = (MapEntry) {
2491         .start = sector_num * BDRV_SECTOR_SIZE,
2492         .length = nb_sectors * BDRV_SECTOR_SIZE,
2493         .data = !!(ret & BDRV_BLOCK_DATA),
2494         .zero = !!(ret & BDRV_BLOCK_ZERO),
2495         .offset = ret & BDRV_BLOCK_OFFSET_MASK,
2496         .has_offset = has_offset,
2497         .depth = depth,
2498         .has_filename = file && has_offset,
2499         .filename = file && has_offset ? file->filename : NULL,
2500     };
2501 
2502     return 0;
2503 }
2504 
2505 static inline bool entry_mergeable(const MapEntry *curr, const MapEntry *next)
2506 {
2507     if (curr->length == 0) {
2508         return false;
2509     }
2510     if (curr->zero != next->zero ||
2511         curr->data != next->data ||
2512         curr->depth != next->depth ||
2513         curr->has_filename != next->has_filename ||
2514         curr->has_offset != next->has_offset) {
2515         return false;
2516     }
2517     if (curr->has_filename && strcmp(curr->filename, next->filename)) {
2518         return false;
2519     }
2520     if (curr->has_offset && curr->offset + curr->length != next->offset) {
2521         return false;
2522     }
2523     return true;
2524 }
2525 
2526 static int img_map(int argc, char **argv)
2527 {
2528     int c;
2529     OutputFormat output_format = OFORMAT_HUMAN;
2530     BlockBackend *blk;
2531     BlockDriverState *bs;
2532     const char *filename, *fmt, *output;
2533     int64_t length;
2534     MapEntry curr = { .length = 0 }, next;
2535     int ret = 0;
2536     bool image_opts = false;
2537 
2538     fmt = NULL;
2539     output = NULL;
2540     for (;;) {
2541         int option_index = 0;
2542         static const struct option long_options[] = {
2543             {"help", no_argument, 0, 'h'},
2544             {"format", required_argument, 0, 'f'},
2545             {"output", required_argument, 0, OPTION_OUTPUT},
2546             {"object", required_argument, 0, OPTION_OBJECT},
2547             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2548             {0, 0, 0, 0}
2549         };
2550         c = getopt_long(argc, argv, "f:h",
2551                         long_options, &option_index);
2552         if (c == -1) {
2553             break;
2554         }
2555         switch (c) {
2556         case '?':
2557         case 'h':
2558             help();
2559             break;
2560         case 'f':
2561             fmt = optarg;
2562             break;
2563         case OPTION_OUTPUT:
2564             output = optarg;
2565             break;
2566         case OPTION_OBJECT: {
2567             QemuOpts *opts;
2568             opts = qemu_opts_parse_noisily(&qemu_object_opts,
2569                                            optarg, true);
2570             if (!opts) {
2571                 return 1;
2572             }
2573         }   break;
2574         case OPTION_IMAGE_OPTS:
2575             image_opts = true;
2576             break;
2577         }
2578     }
2579     if (optind != argc - 1) {
2580         error_exit("Expecting one image file name");
2581     }
2582     filename = argv[optind];
2583 
2584     if (output && !strcmp(output, "json")) {
2585         output_format = OFORMAT_JSON;
2586     } else if (output && !strcmp(output, "human")) {
2587         output_format = OFORMAT_HUMAN;
2588     } else if (output) {
2589         error_report("--output must be used with human or json as argument.");
2590         return 1;
2591     }
2592 
2593     if (qemu_opts_foreach(&qemu_object_opts,
2594                           user_creatable_add_opts_foreach,
2595                           NULL, NULL)) {
2596         return 1;
2597     }
2598 
2599     blk = img_open(image_opts, filename, fmt, 0, false, false);
2600     if (!blk) {
2601         return 1;
2602     }
2603     bs = blk_bs(blk);
2604 
2605     if (output_format == OFORMAT_HUMAN) {
2606         printf("%-16s%-16s%-16s%s\n", "Offset", "Length", "Mapped to", "File");
2607     }
2608 
2609     length = blk_getlength(blk);
2610     while (curr.start + curr.length < length) {
2611         int64_t nsectors_left;
2612         int64_t sector_num;
2613         int n;
2614 
2615         sector_num = (curr.start + curr.length) >> BDRV_SECTOR_BITS;
2616 
2617         /* Probe up to 1 GiB at a time.  */
2618         nsectors_left = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE) - sector_num;
2619         n = MIN(1 << (30 - BDRV_SECTOR_BITS), nsectors_left);
2620         ret = get_block_status(bs, sector_num, n, &next);
2621 
2622         if (ret < 0) {
2623             error_report("Could not read file metadata: %s", strerror(-ret));
2624             goto out;
2625         }
2626 
2627         if (entry_mergeable(&curr, &next)) {
2628             curr.length += next.length;
2629             continue;
2630         }
2631 
2632         if (curr.length > 0) {
2633             dump_map_entry(output_format, &curr, &next);
2634         }
2635         curr = next;
2636     }
2637 
2638     dump_map_entry(output_format, &curr, NULL);
2639 
2640 out:
2641     blk_unref(blk);
2642     return ret < 0;
2643 }
2644 
2645 #define SNAPSHOT_LIST   1
2646 #define SNAPSHOT_CREATE 2
2647 #define SNAPSHOT_APPLY  3
2648 #define SNAPSHOT_DELETE 4
2649 
2650 static int img_snapshot(int argc, char **argv)
2651 {
2652     BlockBackend *blk;
2653     BlockDriverState *bs;
2654     QEMUSnapshotInfo sn;
2655     char *filename, *snapshot_name = NULL;
2656     int c, ret = 0, bdrv_oflags;
2657     int action = 0;
2658     qemu_timeval tv;
2659     bool quiet = false;
2660     Error *err = NULL;
2661     bool image_opts = false;
2662 
2663     bdrv_oflags = BDRV_O_RDWR;
2664     /* Parse commandline parameters */
2665     for(;;) {
2666         static const struct option long_options[] = {
2667             {"help", no_argument, 0, 'h'},
2668             {"object", required_argument, 0, OPTION_OBJECT},
2669             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2670             {0, 0, 0, 0}
2671         };
2672         c = getopt_long(argc, argv, "la:c:d:hq",
2673                         long_options, NULL);
2674         if (c == -1) {
2675             break;
2676         }
2677         switch(c) {
2678         case '?':
2679         case 'h':
2680             help();
2681             return 0;
2682         case 'l':
2683             if (action) {
2684                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
2685                 return 0;
2686             }
2687             action = SNAPSHOT_LIST;
2688             bdrv_oflags &= ~BDRV_O_RDWR; /* no need for RW */
2689             break;
2690         case 'a':
2691             if (action) {
2692                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
2693                 return 0;
2694             }
2695             action = SNAPSHOT_APPLY;
2696             snapshot_name = optarg;
2697             break;
2698         case 'c':
2699             if (action) {
2700                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
2701                 return 0;
2702             }
2703             action = SNAPSHOT_CREATE;
2704             snapshot_name = optarg;
2705             break;
2706         case 'd':
2707             if (action) {
2708                 error_exit("Cannot mix '-l', '-a', '-c', '-d'");
2709                 return 0;
2710             }
2711             action = SNAPSHOT_DELETE;
2712             snapshot_name = optarg;
2713             break;
2714         case 'q':
2715             quiet = true;
2716             break;
2717         case OPTION_OBJECT: {
2718             QemuOpts *opts;
2719             opts = qemu_opts_parse_noisily(&qemu_object_opts,
2720                                            optarg, true);
2721             if (!opts) {
2722                 return 1;
2723             }
2724         }   break;
2725         case OPTION_IMAGE_OPTS:
2726             image_opts = true;
2727             break;
2728         }
2729     }
2730 
2731     if (optind != argc - 1) {
2732         error_exit("Expecting one image file name");
2733     }
2734     filename = argv[optind++];
2735 
2736     if (qemu_opts_foreach(&qemu_object_opts,
2737                           user_creatable_add_opts_foreach,
2738                           NULL, NULL)) {
2739         return 1;
2740     }
2741 
2742     /* Open the image */
2743     blk = img_open(image_opts, filename, NULL, bdrv_oflags, false, quiet);
2744     if (!blk) {
2745         return 1;
2746     }
2747     bs = blk_bs(blk);
2748 
2749     /* Perform the requested action */
2750     switch(action) {
2751     case SNAPSHOT_LIST:
2752         dump_snapshots(bs);
2753         break;
2754 
2755     case SNAPSHOT_CREATE:
2756         memset(&sn, 0, sizeof(sn));
2757         pstrcpy(sn.name, sizeof(sn.name), snapshot_name);
2758 
2759         qemu_gettimeofday(&tv);
2760         sn.date_sec = tv.tv_sec;
2761         sn.date_nsec = tv.tv_usec * 1000;
2762 
2763         ret = bdrv_snapshot_create(bs, &sn);
2764         if (ret) {
2765             error_report("Could not create snapshot '%s': %d (%s)",
2766                 snapshot_name, ret, strerror(-ret));
2767         }
2768         break;
2769 
2770     case SNAPSHOT_APPLY:
2771         ret = bdrv_snapshot_goto(bs, snapshot_name);
2772         if (ret) {
2773             error_report("Could not apply snapshot '%s': %d (%s)",
2774                 snapshot_name, ret, strerror(-ret));
2775         }
2776         break;
2777 
2778     case SNAPSHOT_DELETE:
2779         bdrv_snapshot_delete_by_id_or_name(bs, snapshot_name, &err);
2780         if (err) {
2781             error_reportf_err(err, "Could not delete snapshot '%s': ",
2782                               snapshot_name);
2783             ret = 1;
2784         }
2785         break;
2786     }
2787 
2788     /* Cleanup */
2789     blk_unref(blk);
2790     if (ret) {
2791         return 1;
2792     }
2793     return 0;
2794 }
2795 
2796 static int img_rebase(int argc, char **argv)
2797 {
2798     BlockBackend *blk = NULL, *blk_old_backing = NULL, *blk_new_backing = NULL;
2799     uint8_t *buf_old = NULL;
2800     uint8_t *buf_new = NULL;
2801     BlockDriverState *bs = NULL;
2802     char *filename;
2803     const char *fmt, *cache, *src_cache, *out_basefmt, *out_baseimg;
2804     int c, flags, src_flags, ret;
2805     bool writethrough, src_writethrough;
2806     int unsafe = 0;
2807     int progress = 0;
2808     bool quiet = false;
2809     Error *local_err = NULL;
2810     bool image_opts = false;
2811 
2812     /* Parse commandline parameters */
2813     fmt = NULL;
2814     cache = BDRV_DEFAULT_CACHE;
2815     src_cache = BDRV_DEFAULT_CACHE;
2816     out_baseimg = NULL;
2817     out_basefmt = NULL;
2818     for(;;) {
2819         static const struct option long_options[] = {
2820             {"help", no_argument, 0, 'h'},
2821             {"object", required_argument, 0, OPTION_OBJECT},
2822             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2823             {0, 0, 0, 0}
2824         };
2825         c = getopt_long(argc, argv, "hf:F:b:upt:T:q",
2826                         long_options, NULL);
2827         if (c == -1) {
2828             break;
2829         }
2830         switch(c) {
2831         case '?':
2832         case 'h':
2833             help();
2834             return 0;
2835         case 'f':
2836             fmt = optarg;
2837             break;
2838         case 'F':
2839             out_basefmt = optarg;
2840             break;
2841         case 'b':
2842             out_baseimg = optarg;
2843             break;
2844         case 'u':
2845             unsafe = 1;
2846             break;
2847         case 'p':
2848             progress = 1;
2849             break;
2850         case 't':
2851             cache = optarg;
2852             break;
2853         case 'T':
2854             src_cache = optarg;
2855             break;
2856         case 'q':
2857             quiet = true;
2858             break;
2859         case OPTION_OBJECT: {
2860             QemuOpts *opts;
2861             opts = qemu_opts_parse_noisily(&qemu_object_opts,
2862                                            optarg, true);
2863             if (!opts) {
2864                 return 1;
2865             }
2866         }   break;
2867         case OPTION_IMAGE_OPTS:
2868             image_opts = true;
2869             break;
2870         }
2871     }
2872 
2873     if (quiet) {
2874         progress = 0;
2875     }
2876 
2877     if (optind != argc - 1) {
2878         error_exit("Expecting one image file name");
2879     }
2880     if (!unsafe && !out_baseimg) {
2881         error_exit("Must specify backing file (-b) or use unsafe mode (-u)");
2882     }
2883     filename = argv[optind++];
2884 
2885     if (qemu_opts_foreach(&qemu_object_opts,
2886                           user_creatable_add_opts_foreach,
2887                           NULL, NULL)) {
2888         return 1;
2889     }
2890 
2891     qemu_progress_init(progress, 2.0);
2892     qemu_progress_print(0, 100);
2893 
2894     flags = BDRV_O_RDWR | (unsafe ? BDRV_O_NO_BACKING : 0);
2895     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
2896     if (ret < 0) {
2897         error_report("Invalid cache option: %s", cache);
2898         goto out;
2899     }
2900 
2901     src_flags = 0;
2902     ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
2903     if (ret < 0) {
2904         error_report("Invalid source cache option: %s", src_cache);
2905         goto out;
2906     }
2907 
2908     /* The source files are opened read-only, don't care about WCE */
2909     assert((src_flags & BDRV_O_RDWR) == 0);
2910     (void) src_writethrough;
2911 
2912     /*
2913      * Open the images.
2914      *
2915      * Ignore the old backing file for unsafe rebase in case we want to correct
2916      * the reference to a renamed or moved backing file.
2917      */
2918     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet);
2919     if (!blk) {
2920         ret = -1;
2921         goto out;
2922     }
2923     bs = blk_bs(blk);
2924 
2925     if (out_basefmt != NULL) {
2926         if (bdrv_find_format(out_basefmt) == NULL) {
2927             error_report("Invalid format name: '%s'", out_basefmt);
2928             ret = -1;
2929             goto out;
2930         }
2931     }
2932 
2933     /* For safe rebasing we need to compare old and new backing file */
2934     if (!unsafe) {
2935         char backing_name[PATH_MAX];
2936         QDict *options = NULL;
2937 
2938         if (bs->backing_format[0] != '\0') {
2939             options = qdict_new();
2940             qdict_put(options, "driver", qstring_from_str(bs->backing_format));
2941         }
2942 
2943         bdrv_get_backing_filename(bs, backing_name, sizeof(backing_name));
2944         blk_old_backing = blk_new_open(backing_name, NULL,
2945                                        options, src_flags, &local_err);
2946         if (!blk_old_backing) {
2947             error_reportf_err(local_err,
2948                               "Could not open old backing file '%s': ",
2949                               backing_name);
2950             goto out;
2951         }
2952 
2953         if (out_baseimg[0]) {
2954             if (out_basefmt) {
2955                 options = qdict_new();
2956                 qdict_put(options, "driver", qstring_from_str(out_basefmt));
2957             } else {
2958                 options = NULL;
2959             }
2960 
2961             blk_new_backing = blk_new_open(out_baseimg, NULL,
2962                                            options, src_flags, &local_err);
2963             if (!blk_new_backing) {
2964                 error_reportf_err(local_err,
2965                                   "Could not open new backing file '%s': ",
2966                                   out_baseimg);
2967                 goto out;
2968             }
2969         }
2970     }
2971 
2972     /*
2973      * Check each unallocated cluster in the COW file. If it is unallocated,
2974      * accesses go to the backing file. We must therefore compare this cluster
2975      * in the old and new backing file, and if they differ we need to copy it
2976      * from the old backing file into the COW file.
2977      *
2978      * If qemu-img crashes during this step, no harm is done. The content of
2979      * the image is the same as the original one at any time.
2980      */
2981     if (!unsafe) {
2982         int64_t num_sectors;
2983         int64_t old_backing_num_sectors;
2984         int64_t new_backing_num_sectors = 0;
2985         uint64_t sector;
2986         int n;
2987         float local_progress = 0;
2988 
2989         buf_old = blk_blockalign(blk, IO_BUF_SIZE);
2990         buf_new = blk_blockalign(blk, IO_BUF_SIZE);
2991 
2992         num_sectors = blk_nb_sectors(blk);
2993         if (num_sectors < 0) {
2994             error_report("Could not get size of '%s': %s",
2995                          filename, strerror(-num_sectors));
2996             ret = -1;
2997             goto out;
2998         }
2999         old_backing_num_sectors = blk_nb_sectors(blk_old_backing);
3000         if (old_backing_num_sectors < 0) {
3001             char backing_name[PATH_MAX];
3002 
3003             bdrv_get_backing_filename(bs, backing_name, sizeof(backing_name));
3004             error_report("Could not get size of '%s': %s",
3005                          backing_name, strerror(-old_backing_num_sectors));
3006             ret = -1;
3007             goto out;
3008         }
3009         if (blk_new_backing) {
3010             new_backing_num_sectors = blk_nb_sectors(blk_new_backing);
3011             if (new_backing_num_sectors < 0) {
3012                 error_report("Could not get size of '%s': %s",
3013                              out_baseimg, strerror(-new_backing_num_sectors));
3014                 ret = -1;
3015                 goto out;
3016             }
3017         }
3018 
3019         if (num_sectors != 0) {
3020             local_progress = (float)100 /
3021                 (num_sectors / MIN(num_sectors, IO_BUF_SIZE / 512));
3022         }
3023 
3024         for (sector = 0; sector < num_sectors; sector += n) {
3025 
3026             /* How many sectors can we handle with the next read? */
3027             if (sector + (IO_BUF_SIZE / 512) <= num_sectors) {
3028                 n = (IO_BUF_SIZE / 512);
3029             } else {
3030                 n = num_sectors - sector;
3031             }
3032 
3033             /* If the cluster is allocated, we don't need to take action */
3034             ret = bdrv_is_allocated(bs, sector, n, &n);
3035             if (ret < 0) {
3036                 error_report("error while reading image metadata: %s",
3037                              strerror(-ret));
3038                 goto out;
3039             }
3040             if (ret) {
3041                 continue;
3042             }
3043 
3044             /*
3045              * Read old and new backing file and take into consideration that
3046              * backing files may be smaller than the COW image.
3047              */
3048             if (sector >= old_backing_num_sectors) {
3049                 memset(buf_old, 0, n * BDRV_SECTOR_SIZE);
3050             } else {
3051                 if (sector + n > old_backing_num_sectors) {
3052                     n = old_backing_num_sectors - sector;
3053                 }
3054 
3055                 ret = blk_pread(blk_old_backing, sector << BDRV_SECTOR_BITS,
3056                                 buf_old, n << BDRV_SECTOR_BITS);
3057                 if (ret < 0) {
3058                     error_report("error while reading from old backing file");
3059                     goto out;
3060                 }
3061             }
3062 
3063             if (sector >= new_backing_num_sectors || !blk_new_backing) {
3064                 memset(buf_new, 0, n * BDRV_SECTOR_SIZE);
3065             } else {
3066                 if (sector + n > new_backing_num_sectors) {
3067                     n = new_backing_num_sectors - sector;
3068                 }
3069 
3070                 ret = blk_pread(blk_new_backing, sector << BDRV_SECTOR_BITS,
3071                                 buf_new, n << BDRV_SECTOR_BITS);
3072                 if (ret < 0) {
3073                     error_report("error while reading from new backing file");
3074                     goto out;
3075                 }
3076             }
3077 
3078             /* If they differ, we need to write to the COW file */
3079             uint64_t written = 0;
3080 
3081             while (written < n) {
3082                 int pnum;
3083 
3084                 if (compare_sectors(buf_old + written * 512,
3085                     buf_new + written * 512, n - written, &pnum))
3086                 {
3087                     ret = blk_pwrite(blk,
3088                                      (sector + written) << BDRV_SECTOR_BITS,
3089                                      buf_old + written * 512,
3090                                      pnum << BDRV_SECTOR_BITS, 0);
3091                     if (ret < 0) {
3092                         error_report("Error while writing to COW image: %s",
3093                             strerror(-ret));
3094                         goto out;
3095                     }
3096                 }
3097 
3098                 written += pnum;
3099             }
3100             qemu_progress_print(local_progress, 100);
3101         }
3102     }
3103 
3104     /*
3105      * Change the backing file. All clusters that are different from the old
3106      * backing file are overwritten in the COW file now, so the visible content
3107      * doesn't change when we switch the backing file.
3108      */
3109     if (out_baseimg && *out_baseimg) {
3110         ret = bdrv_change_backing_file(bs, out_baseimg, out_basefmt);
3111     } else {
3112         ret = bdrv_change_backing_file(bs, NULL, NULL);
3113     }
3114 
3115     if (ret == -ENOSPC) {
3116         error_report("Could not change the backing file to '%s': No "
3117                      "space left in the file header", out_baseimg);
3118     } else if (ret < 0) {
3119         error_report("Could not change the backing file to '%s': %s",
3120             out_baseimg, strerror(-ret));
3121     }
3122 
3123     qemu_progress_print(100, 0);
3124     /*
3125      * TODO At this point it is possible to check if any clusters that are
3126      * allocated in the COW file are the same in the backing file. If so, they
3127      * could be dropped from the COW file. Don't do this before switching the
3128      * backing file, in case of a crash this would lead to corruption.
3129      */
3130 out:
3131     qemu_progress_end();
3132     /* Cleanup */
3133     if (!unsafe) {
3134         blk_unref(blk_old_backing);
3135         blk_unref(blk_new_backing);
3136     }
3137     qemu_vfree(buf_old);
3138     qemu_vfree(buf_new);
3139 
3140     blk_unref(blk);
3141     if (ret) {
3142         return 1;
3143     }
3144     return 0;
3145 }
3146 
3147 static int img_resize(int argc, char **argv)
3148 {
3149     Error *err = NULL;
3150     int c, ret, relative;
3151     const char *filename, *fmt, *size;
3152     int64_t n, total_size;
3153     bool quiet = false;
3154     BlockBackend *blk = NULL;
3155     QemuOpts *param;
3156 
3157     static QemuOptsList resize_options = {
3158         .name = "resize_options",
3159         .head = QTAILQ_HEAD_INITIALIZER(resize_options.head),
3160         .desc = {
3161             {
3162                 .name = BLOCK_OPT_SIZE,
3163                 .type = QEMU_OPT_SIZE,
3164                 .help = "Virtual disk size"
3165             }, {
3166                 /* end of list */
3167             }
3168         },
3169     };
3170     bool image_opts = false;
3171 
3172     /* Remove size from argv manually so that negative numbers are not treated
3173      * as options by getopt. */
3174     if (argc < 3) {
3175         error_exit("Not enough arguments");
3176         return 1;
3177     }
3178 
3179     size = argv[--argc];
3180 
3181     /* Parse getopt arguments */
3182     fmt = NULL;
3183     for(;;) {
3184         static const struct option long_options[] = {
3185             {"help", no_argument, 0, 'h'},
3186             {"object", required_argument, 0, OPTION_OBJECT},
3187             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3188             {0, 0, 0, 0}
3189         };
3190         c = getopt_long(argc, argv, "f:hq",
3191                         long_options, NULL);
3192         if (c == -1) {
3193             break;
3194         }
3195         switch(c) {
3196         case '?':
3197         case 'h':
3198             help();
3199             break;
3200         case 'f':
3201             fmt = optarg;
3202             break;
3203         case 'q':
3204             quiet = true;
3205             break;
3206         case OPTION_OBJECT: {
3207             QemuOpts *opts;
3208             opts = qemu_opts_parse_noisily(&qemu_object_opts,
3209                                            optarg, true);
3210             if (!opts) {
3211                 return 1;
3212             }
3213         }   break;
3214         case OPTION_IMAGE_OPTS:
3215             image_opts = true;
3216             break;
3217         }
3218     }
3219     if (optind != argc - 1) {
3220         error_exit("Expecting one image file name");
3221     }
3222     filename = argv[optind++];
3223 
3224     if (qemu_opts_foreach(&qemu_object_opts,
3225                           user_creatable_add_opts_foreach,
3226                           NULL, NULL)) {
3227         return 1;
3228     }
3229 
3230     /* Choose grow, shrink, or absolute resize mode */
3231     switch (size[0]) {
3232     case '+':
3233         relative = 1;
3234         size++;
3235         break;
3236     case '-':
3237         relative = -1;
3238         size++;
3239         break;
3240     default:
3241         relative = 0;
3242         break;
3243     }
3244 
3245     /* Parse size */
3246     param = qemu_opts_create(&resize_options, NULL, 0, &error_abort);
3247     qemu_opt_set(param, BLOCK_OPT_SIZE, size, &err);
3248     if (err) {
3249         error_report_err(err);
3250         ret = -1;
3251         qemu_opts_del(param);
3252         goto out;
3253     }
3254     n = qemu_opt_get_size(param, BLOCK_OPT_SIZE, 0);
3255     qemu_opts_del(param);
3256 
3257     blk = img_open(image_opts, filename, fmt,
3258                    BDRV_O_RDWR, false, quiet);
3259     if (!blk) {
3260         ret = -1;
3261         goto out;
3262     }
3263 
3264     if (relative) {
3265         total_size = blk_getlength(blk) + n * relative;
3266     } else {
3267         total_size = n;
3268     }
3269     if (total_size <= 0) {
3270         error_report("New image size must be positive");
3271         ret = -1;
3272         goto out;
3273     }
3274 
3275     ret = blk_truncate(blk, total_size);
3276     switch (ret) {
3277     case 0:
3278         qprintf(quiet, "Image resized.\n");
3279         break;
3280     case -ENOTSUP:
3281         error_report("This image does not support resize");
3282         break;
3283     case -EACCES:
3284         error_report("Image is read-only");
3285         break;
3286     default:
3287         error_report("Error resizing image (%d)", -ret);
3288         break;
3289     }
3290 out:
3291     blk_unref(blk);
3292     if (ret) {
3293         return 1;
3294     }
3295     return 0;
3296 }
3297 
3298 static void amend_status_cb(BlockDriverState *bs,
3299                             int64_t offset, int64_t total_work_size,
3300                             void *opaque)
3301 {
3302     qemu_progress_print(100.f * offset / total_work_size, 0);
3303 }
3304 
3305 static int img_amend(int argc, char **argv)
3306 {
3307     Error *err = NULL;
3308     int c, ret = 0;
3309     char *options = NULL;
3310     QemuOptsList *create_opts = NULL;
3311     QemuOpts *opts = NULL;
3312     const char *fmt = NULL, *filename, *cache;
3313     int flags;
3314     bool writethrough;
3315     bool quiet = false, progress = false;
3316     BlockBackend *blk = NULL;
3317     BlockDriverState *bs = NULL;
3318     bool image_opts = false;
3319 
3320     cache = BDRV_DEFAULT_CACHE;
3321     for (;;) {
3322         static const struct option long_options[] = {
3323             {"help", no_argument, 0, 'h'},
3324             {"object", required_argument, 0, OPTION_OBJECT},
3325             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3326             {0, 0, 0, 0}
3327         };
3328         c = getopt_long(argc, argv, "ho:f:t:pq",
3329                         long_options, NULL);
3330         if (c == -1) {
3331             break;
3332         }
3333 
3334         switch (c) {
3335             case 'h':
3336             case '?':
3337                 help();
3338                 break;
3339             case 'o':
3340                 if (!is_valid_option_list(optarg)) {
3341                     error_report("Invalid option list: %s", optarg);
3342                     ret = -1;
3343                     goto out_no_progress;
3344                 }
3345                 if (!options) {
3346                     options = g_strdup(optarg);
3347                 } else {
3348                     char *old_options = options;
3349                     options = g_strdup_printf("%s,%s", options, optarg);
3350                     g_free(old_options);
3351                 }
3352                 break;
3353             case 'f':
3354                 fmt = optarg;
3355                 break;
3356             case 't':
3357                 cache = optarg;
3358                 break;
3359             case 'p':
3360                 progress = true;
3361                 break;
3362             case 'q':
3363                 quiet = true;
3364                 break;
3365             case OPTION_OBJECT:
3366                 opts = qemu_opts_parse_noisily(&qemu_object_opts,
3367                                                optarg, true);
3368                 if (!opts) {
3369                     ret = -1;
3370                     goto out_no_progress;
3371                 }
3372                 break;
3373             case OPTION_IMAGE_OPTS:
3374                 image_opts = true;
3375                 break;
3376         }
3377     }
3378 
3379     if (!options) {
3380         error_exit("Must specify options (-o)");
3381     }
3382 
3383     if (qemu_opts_foreach(&qemu_object_opts,
3384                           user_creatable_add_opts_foreach,
3385                           NULL, NULL)) {
3386         ret = -1;
3387         goto out_no_progress;
3388     }
3389 
3390     if (quiet) {
3391         progress = false;
3392     }
3393     qemu_progress_init(progress, 1.0);
3394 
3395     filename = (optind == argc - 1) ? argv[argc - 1] : NULL;
3396     if (fmt && has_help_option(options)) {
3397         /* If a format is explicitly specified (and possibly no filename is
3398          * given), print option help here */
3399         ret = print_block_option_help(filename, fmt);
3400         goto out;
3401     }
3402 
3403     if (optind != argc - 1) {
3404         error_report("Expecting one image file name");
3405         ret = -1;
3406         goto out;
3407     }
3408 
3409     flags = BDRV_O_RDWR;
3410     ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
3411     if (ret < 0) {
3412         error_report("Invalid cache option: %s", cache);
3413         goto out;
3414     }
3415 
3416     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet);
3417     if (!blk) {
3418         ret = -1;
3419         goto out;
3420     }
3421     bs = blk_bs(blk);
3422 
3423     fmt = bs->drv->format_name;
3424 
3425     if (has_help_option(options)) {
3426         /* If the format was auto-detected, print option help here */
3427         ret = print_block_option_help(filename, fmt);
3428         goto out;
3429     }
3430 
3431     if (!bs->drv->create_opts) {
3432         error_report("Format driver '%s' does not support any options to amend",
3433                      fmt);
3434         ret = -1;
3435         goto out;
3436     }
3437 
3438     create_opts = qemu_opts_append(create_opts, bs->drv->create_opts);
3439     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
3440     if (options) {
3441         qemu_opts_do_parse(opts, options, NULL, &err);
3442         if (err) {
3443             error_report_err(err);
3444             ret = -1;
3445             goto out;
3446         }
3447     }
3448 
3449     /* In case the driver does not call amend_status_cb() */
3450     qemu_progress_print(0.f, 0);
3451     ret = bdrv_amend_options(bs, opts, &amend_status_cb, NULL);
3452     qemu_progress_print(100.f, 0);
3453     if (ret < 0) {
3454         error_report("Error while amending options: %s", strerror(-ret));
3455         goto out;
3456     }
3457 
3458 out:
3459     qemu_progress_end();
3460 
3461 out_no_progress:
3462     blk_unref(blk);
3463     qemu_opts_del(opts);
3464     qemu_opts_free(create_opts);
3465     g_free(options);
3466 
3467     if (ret) {
3468         return 1;
3469     }
3470     return 0;
3471 }
3472 
3473 typedef struct BenchData {
3474     BlockBackend *blk;
3475     uint64_t image_size;
3476     bool write;
3477     int bufsize;
3478     int step;
3479     int nrreq;
3480     int n;
3481     int flush_interval;
3482     bool drain_on_flush;
3483     uint8_t *buf;
3484     QEMUIOVector *qiov;
3485 
3486     int in_flight;
3487     bool in_flush;
3488     uint64_t offset;
3489 } BenchData;
3490 
3491 static void bench_undrained_flush_cb(void *opaque, int ret)
3492 {
3493     if (ret < 0) {
3494         error_report("Failed flush request: %s\n", strerror(-ret));
3495         exit(EXIT_FAILURE);
3496     }
3497 }
3498 
3499 static void bench_cb(void *opaque, int ret)
3500 {
3501     BenchData *b = opaque;
3502     BlockAIOCB *acb;
3503 
3504     if (ret < 0) {
3505         error_report("Failed request: %s\n", strerror(-ret));
3506         exit(EXIT_FAILURE);
3507     }
3508 
3509     if (b->in_flush) {
3510         /* Just finished a flush with drained queue: Start next requests */
3511         assert(b->in_flight == 0);
3512         b->in_flush = false;
3513     } else if (b->in_flight > 0) {
3514         int remaining = b->n - b->in_flight;
3515 
3516         b->n--;
3517         b->in_flight--;
3518 
3519         /* Time for flush? Drain queue if requested, then flush */
3520         if (b->flush_interval && remaining % b->flush_interval == 0) {
3521             if (!b->in_flight || !b->drain_on_flush) {
3522                 BlockCompletionFunc *cb;
3523 
3524                 if (b->drain_on_flush) {
3525                     b->in_flush = true;
3526                     cb = bench_cb;
3527                 } else {
3528                     cb = bench_undrained_flush_cb;
3529                 }
3530 
3531                 acb = blk_aio_flush(b->blk, cb, b);
3532                 if (!acb) {
3533                     error_report("Failed to issue flush request");
3534                     exit(EXIT_FAILURE);
3535                 }
3536             }
3537             if (b->drain_on_flush) {
3538                 return;
3539             }
3540         }
3541     }
3542 
3543     while (b->n > b->in_flight && b->in_flight < b->nrreq) {
3544         if (b->write) {
3545             acb = blk_aio_pwritev(b->blk, b->offset, b->qiov, 0,
3546                                   bench_cb, b);
3547         } else {
3548             acb = blk_aio_preadv(b->blk, b->offset, b->qiov, 0,
3549                                  bench_cb, b);
3550         }
3551         if (!acb) {
3552             error_report("Failed to issue request");
3553             exit(EXIT_FAILURE);
3554         }
3555         b->in_flight++;
3556         b->offset += b->step;
3557         b->offset %= b->image_size;
3558     }
3559 }
3560 
3561 static int img_bench(int argc, char **argv)
3562 {
3563     int c, ret = 0;
3564     const char *fmt = NULL, *filename;
3565     bool quiet = false;
3566     bool image_opts = false;
3567     bool is_write = false;
3568     int count = 75000;
3569     int depth = 64;
3570     int64_t offset = 0;
3571     size_t bufsize = 4096;
3572     int pattern = 0;
3573     size_t step = 0;
3574     int flush_interval = 0;
3575     bool drain_on_flush = true;
3576     int64_t image_size;
3577     BlockBackend *blk = NULL;
3578     BenchData data = {};
3579     int flags = 0;
3580     bool writethrough = false;
3581     struct timeval t1, t2;
3582     int i;
3583 
3584     for (;;) {
3585         static const struct option long_options[] = {
3586             {"help", no_argument, 0, 'h'},
3587             {"flush-interval", required_argument, 0, OPTION_FLUSH_INTERVAL},
3588             {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3589             {"pattern", required_argument, 0, OPTION_PATTERN},
3590             {"no-drain", no_argument, 0, OPTION_NO_DRAIN},
3591             {0, 0, 0, 0}
3592         };
3593         c = getopt_long(argc, argv, "hc:d:f:no:qs:S:t:w", long_options, NULL);
3594         if (c == -1) {
3595             break;
3596         }
3597 
3598         switch (c) {
3599         case 'h':
3600         case '?':
3601             help();
3602             break;
3603         case 'c':
3604         {
3605             char *end;
3606             errno = 0;
3607             count = strtoul(optarg, &end, 0);
3608             if (errno || *end || count > INT_MAX) {
3609                 error_report("Invalid request count specified");
3610                 return 1;
3611             }
3612             break;
3613         }
3614         case 'd':
3615         {
3616             char *end;
3617             errno = 0;
3618             depth = strtoul(optarg, &end, 0);
3619             if (errno || *end || depth > INT_MAX) {
3620                 error_report("Invalid queue depth specified");
3621                 return 1;
3622             }
3623             break;
3624         }
3625         case 'f':
3626             fmt = optarg;
3627             break;
3628         case 'n':
3629             flags |= BDRV_O_NATIVE_AIO;
3630             break;
3631         case 'o':
3632         {
3633             char *end;
3634             errno = 0;
3635             offset = qemu_strtosz_suffix(optarg, &end,
3636                                          QEMU_STRTOSZ_DEFSUFFIX_B);
3637             if (offset < 0|| *end) {
3638                 error_report("Invalid offset specified");
3639                 return 1;
3640             }
3641             break;
3642         }
3643             break;
3644         case 'q':
3645             quiet = true;
3646             break;
3647         case 's':
3648         {
3649             int64_t sval;
3650             char *end;
3651 
3652             sval = qemu_strtosz_suffix(optarg, &end, QEMU_STRTOSZ_DEFSUFFIX_B);
3653             if (sval < 0 || sval > INT_MAX || *end) {
3654                 error_report("Invalid buffer size specified");
3655                 return 1;
3656             }
3657 
3658             bufsize = sval;
3659             break;
3660         }
3661         case 'S':
3662         {
3663             int64_t sval;
3664             char *end;
3665 
3666             sval = qemu_strtosz_suffix(optarg, &end, QEMU_STRTOSZ_DEFSUFFIX_B);
3667             if (sval < 0 || sval > INT_MAX || *end) {
3668                 error_report("Invalid step size specified");
3669                 return 1;
3670             }
3671 
3672             step = sval;
3673             break;
3674         }
3675         case 't':
3676             ret = bdrv_parse_cache_mode(optarg, &flags, &writethrough);
3677             if (ret < 0) {
3678                 error_report("Invalid cache mode");
3679                 ret = -1;
3680                 goto out;
3681             }
3682             break;
3683         case 'w':
3684             flags |= BDRV_O_RDWR;
3685             is_write = true;
3686             break;
3687         case OPTION_PATTERN:
3688         {
3689             char *end;
3690             errno = 0;
3691             pattern = strtoul(optarg, &end, 0);
3692             if (errno || *end || pattern > 0xff) {
3693                 error_report("Invalid pattern byte specified");
3694                 return 1;
3695             }
3696             break;
3697         }
3698         case OPTION_FLUSH_INTERVAL:
3699         {
3700             char *end;
3701             errno = 0;
3702             flush_interval = strtoul(optarg, &end, 0);
3703             if (errno || *end || flush_interval > INT_MAX) {
3704                 error_report("Invalid flush interval specified");
3705                 return 1;
3706             }
3707             break;
3708         }
3709         case OPTION_NO_DRAIN:
3710             drain_on_flush = false;
3711             break;
3712         case OPTION_IMAGE_OPTS:
3713             image_opts = true;
3714             break;
3715         }
3716     }
3717 
3718     if (optind != argc - 1) {
3719         error_exit("Expecting one image file name");
3720     }
3721     filename = argv[argc - 1];
3722 
3723     if (!is_write && flush_interval) {
3724         error_report("--flush-interval is only available in write tests");
3725         ret = -1;
3726         goto out;
3727     }
3728     if (flush_interval && flush_interval < depth) {
3729         error_report("Flush interval can't be smaller than depth");
3730         ret = -1;
3731         goto out;
3732     }
3733 
3734     blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet);
3735     if (!blk) {
3736         ret = -1;
3737         goto out;
3738     }
3739 
3740     image_size = blk_getlength(blk);
3741     if (image_size < 0) {
3742         ret = image_size;
3743         goto out;
3744     }
3745 
3746     data = (BenchData) {
3747         .blk            = blk,
3748         .image_size     = image_size,
3749         .bufsize        = bufsize,
3750         .step           = step ?: bufsize,
3751         .nrreq          = depth,
3752         .n              = count,
3753         .offset         = offset,
3754         .write          = is_write,
3755         .flush_interval = flush_interval,
3756         .drain_on_flush = drain_on_flush,
3757     };
3758     printf("Sending %d %s requests, %d bytes each, %d in parallel "
3759            "(starting at offset %" PRId64 ", step size %d)\n",
3760            data.n, data.write ? "write" : "read", data.bufsize, data.nrreq,
3761            data.offset, data.step);
3762     if (flush_interval) {
3763         printf("Sending flush every %d requests\n", flush_interval);
3764     }
3765 
3766     data.buf = blk_blockalign(blk, data.nrreq * data.bufsize);
3767     memset(data.buf, pattern, data.nrreq * data.bufsize);
3768 
3769     data.qiov = g_new(QEMUIOVector, data.nrreq);
3770     for (i = 0; i < data.nrreq; i++) {
3771         qemu_iovec_init(&data.qiov[i], 1);
3772         qemu_iovec_add(&data.qiov[i],
3773                        data.buf + i * data.bufsize, data.bufsize);
3774     }
3775 
3776     gettimeofday(&t1, NULL);
3777     bench_cb(&data, 0);
3778 
3779     while (data.n > 0) {
3780         main_loop_wait(false);
3781     }
3782     gettimeofday(&t2, NULL);
3783 
3784     printf("Run completed in %3.3f seconds.\n",
3785            (t2.tv_sec - t1.tv_sec)
3786            + ((double)(t2.tv_usec - t1.tv_usec) / 1000000));
3787 
3788 out:
3789     qemu_vfree(data.buf);
3790     blk_unref(blk);
3791 
3792     if (ret) {
3793         return 1;
3794     }
3795     return 0;
3796 }
3797 
3798 
3799 static const img_cmd_t img_cmds[] = {
3800 #define DEF(option, callback, arg_string)        \
3801     { option, callback },
3802 #include "qemu-img-cmds.h"
3803 #undef DEF
3804 #undef GEN_DOCS
3805     { NULL, NULL, },
3806 };
3807 
3808 int main(int argc, char **argv)
3809 {
3810     const img_cmd_t *cmd;
3811     const char *cmdname;
3812     Error *local_error = NULL;
3813     char *trace_file = NULL;
3814     int c;
3815     static const struct option long_options[] = {
3816         {"help", no_argument, 0, 'h'},
3817         {"version", no_argument, 0, 'V'},
3818         {"trace", required_argument, NULL, 'T'},
3819         {0, 0, 0, 0}
3820     };
3821 
3822 #ifdef CONFIG_POSIX
3823     signal(SIGPIPE, SIG_IGN);
3824 #endif
3825 
3826     error_set_progname(argv[0]);
3827     qemu_init_exec_dir(argv[0]);
3828 
3829     if (qemu_init_main_loop(&local_error)) {
3830         error_report_err(local_error);
3831         exit(EXIT_FAILURE);
3832     }
3833 
3834     qcrypto_init(&error_fatal);
3835 
3836     module_call_init(MODULE_INIT_QOM);
3837     bdrv_init();
3838     if (argc < 2) {
3839         error_exit("Not enough arguments");
3840     }
3841 
3842     qemu_add_opts(&qemu_object_opts);
3843     qemu_add_opts(&qemu_source_opts);
3844     qemu_add_opts(&qemu_trace_opts);
3845 
3846     while ((c = getopt_long(argc, argv, "+hVT:", long_options, NULL)) != -1) {
3847         switch (c) {
3848         case 'h':
3849             help();
3850             return 0;
3851         case 'V':
3852             printf(QEMU_IMG_VERSION);
3853             return 0;
3854         case 'T':
3855             g_free(trace_file);
3856             trace_file = trace_opt_parse(optarg);
3857             break;
3858         }
3859     }
3860 
3861     cmdname = argv[optind];
3862 
3863     /* reset getopt_long scanning */
3864     argc -= optind;
3865     if (argc < 1) {
3866         return 0;
3867     }
3868     argv += optind;
3869     optind = 1;
3870 
3871     if (!trace_init_backends()) {
3872         exit(1);
3873     }
3874     trace_init_file(trace_file);
3875     qemu_set_log(LOG_TRACE);
3876 
3877     /* find the command */
3878     for (cmd = img_cmds; cmd->name != NULL; cmd++) {
3879         if (!strcmp(cmdname, cmd->name)) {
3880             return cmd->handler(argc, argv);
3881         }
3882     }
3883 
3884     /* not found */
3885     error_exit("Command not found: %s", cmdname);
3886 }
3887