master
c 6,049 lines 184 KB
Raw
1 /*
2 * QEMU disk image utility
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include <getopt.h>
27
28 #include "qemu/help-texts.h"
29 #include "qemu/qemu-progress.h"
30 #include "qemu-version.h"
31 #include "qapi/error.h"
32 #include "qapi/qapi-commands-block-core.h"
33 #include "qapi/qapi-visit-block-core.h"
34 #include "qapi/qobject-output-visitor.h"
35 #include "qobject/qjson.h"
36 #include "qobject/qdict.h"
37 #include "qemu/cutils.h"
38 #include "qemu/config-file.h"
39 #include "qemu/option.h"
40 #include "qemu/error-report.h"
41 #include "qemu/log.h"
42 #include "qemu/main-loop.h"
43 #include "qemu/module.h"
44 #include "qemu/sockets.h"
45 #include "qemu/units.h"
46 #include "qemu/memalign.h"
47 #include "qom/object_interfaces.h"
48 #include "system/block-backend.h"
49 #include "block/block_int.h"
50 #include "block/blockjob.h"
51 #include "block/dirty-bitmap.h"
52 #include "block/qapi.h"
53 #include "crypto/init.h"
54 #include "trace/control.h"
55 #include "qemu/throttle.h"
56 #include "block/throttle-groups.h"
57
58 #define QEMU_IMG_VERSION "qemu-img version " QEMU_FULL_VERSION \
59 "\n" QEMU_COPYRIGHT "\n"
60
61 typedef struct img_cmd_t {
62 const char *name;
63 int (*handler)(const struct img_cmd_t *ccmd, int argc, char **argv);
64 const char *description;
65 } img_cmd_t;
66
67 enum {
68 OPTION_OUTPUT = 256,
69 OPTION_BACKING_CHAIN = 257,
70 OPTION_OBJECT = 258,
71 OPTION_IMAGE_OPTS = 259,
72 OPTION_PATTERN = 260,
73 OPTION_FLUSH_INTERVAL = 261,
74 OPTION_NO_DRAIN = 262,
75 OPTION_TARGET_IMAGE_OPTS = 263,
76 OPTION_PREALLOCATION = 265,
77 OPTION_SHRINK = 266,
78 OPTION_SALVAGE = 267,
79 OPTION_TARGET_IS_ZERO = 268,
80 OPTION_ADD = 269,
81 OPTION_REMOVE = 270,
82 OPTION_CLEAR = 271,
83 OPTION_ENABLE = 272,
84 OPTION_DISABLE = 273,
85 OPTION_MERGE = 274,
86 OPTION_BITMAPS = 275,
87 OPTION_FORCE = 276,
88 OPTION_SKIP_BROKEN = 277,
89 OPTION_LIMITS = 278,
90 OPTION_REMOVE_ALL = 279,
91 };
92
93 typedef enum OutputFormat {
94 OFORMAT_JSON,
95 OFORMAT_HUMAN,
96 } OutputFormat;
97
98 /* Default to cache=writeback as data integrity is not important for qemu-img */
99 #define BDRV_DEFAULT_CACHE "writeback"
100
101 static G_NORETURN
102 void tryhelp(const char *argv0)
103 {
104 error_printf("Try '%s --help' for more information\n", argv0);
105 exit(EXIT_FAILURE);
106 }
107
108 static G_NORETURN G_GNUC_PRINTF(2, 3)
109 void error_exit(const char *argv0, const char *fmt, ...)
110 {
111 va_list ap;
112
113 va_start(ap, fmt);
114 error_vreport(fmt, ap);
115 va_end(ap);
116
117 tryhelp(argv0);
118 }
119
120 /*
121 * Print --help output for a command and exit.
122 * @syntax and @description are multi-line with trailing EOL
123 * (to allow easy extending of the text)
124 * @syntax has each subsequent line indented by 8 chars.
125 * @description is indented by 2 chars for argument on each own line,
126 * and with 5 chars for argument description (like -h arg below).
127 */
128 static G_NORETURN
129 void cmd_help(const img_cmd_t *ccmd,
130 const char *syntax, const char *arguments)
131 {
132 printf(
133 "Usage:\n"
134 " %s %s %s\n"
135 "%s.\n"
136 "\n"
137 "Arguments:\n"
138 " -h, --help\n"
139 " print this help and exit\n"
140 "%s\n",
141 "qemu-img", ccmd->name, syntax, ccmd->description, arguments);
142 exit(EXIT_SUCCESS);
143 }
144
145 static OutputFormat parse_output_format(const char *argv0, const char *arg)
146 {
147 if (!strcmp(arg, "json")) {
148 return OFORMAT_JSON;
149 } else if (!strcmp(arg, "human")) {
150 return OFORMAT_HUMAN;
151 } else {
152 error_exit(argv0, "--output expects 'human' or 'json', not '%s'", arg);
153 }
154 }
155
156 /*
157 * Is @list safe for accumulate_options()?
158 * It is when multiple of them can be joined together separated by ','.
159 * To make that work, @list must not start with ',' (or else a
160 * separating ',' preceding it gets escaped), and it must not end with
161 * an odd number of ',' (or else a separating ',' following it gets
162 * escaped), or be empty (or else a separating ',' preceding it can
163 * escape a separating ',' following it).
164 *
165 */
166 static bool is_valid_option_list(const char *list)
167 {
168 size_t len = strlen(list);
169 size_t i;
170
171 if (!list[0] || list[0] == ',') {
172 return false;
173 }
174
175 for (i = len; i > 0 && list[i - 1] == ','; i--) {
176 }
177 if ((len - i) % 2) {
178 return false;
179 }
180
181 return true;
182 }
183
184 static int accumulate_options(char **options, char *list)
185 {
186 char *new_options;
187
188 if (!is_valid_option_list(list)) {
189 error_report("Invalid option list: %s", list);
190 return -1;
191 }
192
193 if (!*options) {
194 *options = g_strdup(list);
195 } else {
196 new_options = g_strdup_printf("%s,%s", *options, list);
197 g_free(*options);
198 *options = new_options;
199 }
200 return 0;
201 }
202
203 static QemuOptsList qemu_source_opts = {
204 .name = "source",
205 .implied_opt_name = "file",
206 .head = QTAILQ_HEAD_INITIALIZER(qemu_source_opts.head),
207 .desc = {
208 { }
209 },
210 };
211
212 static int G_GNUC_PRINTF(2, 3) qprintf(bool quiet, const char *fmt, ...)
213 {
214 int ret = 0;
215 if (!quiet) {
216 va_list args;
217 va_start(args, fmt);
218 ret = vprintf(fmt, args);
219 va_end(args);
220 }
221 return ret;
222 }
223
224
225 static int print_block_option_help(const char *filename, const char *fmt)
226 {
227 BlockDriver *drv, *proto_drv;
228 QemuOptsList *create_opts = NULL;
229 Error *local_err = NULL;
230
231 /* Find driver and parse its options */
232 drv = bdrv_find_format(fmt);
233 if (!drv) {
234 error_report("Unknown file format '%s'", fmt);
235 return 1;
236 }
237
238 if (!drv->create_opts) {
239 error_report("Format driver '%s' does not support image creation", fmt);
240 return 1;
241 }
242
243 create_opts = qemu_opts_append(create_opts, drv->create_opts);
244 if (filename) {
245 proto_drv = bdrv_find_protocol(filename, true, &local_err);
246 if (!proto_drv) {
247 error_report_err(local_err);
248 qemu_opts_free(create_opts);
249 return 1;
250 }
251 if (!proto_drv->create_opts) {
252 error_report("Protocol driver '%s' does not support image creation",
253 proto_drv->format_name);
254 qemu_opts_free(create_opts);
255 return 1;
256 }
257 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
258 }
259
260 if (filename) {
261 printf("Supported options:\n");
262 } else {
263 printf("Supported %s options:\n", fmt);
264 }
265 qemu_opts_print_help(create_opts, false);
266 qemu_opts_free(create_opts);
267
268 if (!filename) {
269 printf("\n"
270 "The protocol level may support further options.\n"
271 "Specify the target filename to include those options.\n");
272 }
273
274 return 0;
275 }
276
277
278 static BlockBackend *img_open_opts(const char *optstr,
279 QemuOpts *opts, int flags, bool writethrough,
280 bool quiet, bool force_share)
281 {
282 QDict *options;
283 Error *local_err = NULL;
284 BlockBackend *blk;
285 options = qemu_opts_to_qdict(opts, NULL);
286 if (force_share) {
287 if (qdict_haskey(options, BDRV_OPT_FORCE_SHARE)
288 && strcmp(qdict_get_str(options, BDRV_OPT_FORCE_SHARE), "on")) {
289 error_report("--force-share/-U conflicts with image options");
290 qobject_unref(options);
291 return NULL;
292 }
293 qdict_put_str(options, BDRV_OPT_FORCE_SHARE, "on");
294 }
295 blk = blk_new_open(NULL, NULL, options, flags, &local_err);
296 if (!blk) {
297 error_reportf_err(local_err, "Could not open '%s': ", optstr);
298 return NULL;
299 }
300 blk_set_enable_write_cache(blk, !writethrough);
301
302 return blk;
303 }
304
305 static BlockBackend *img_open_file(const char *filename,
306 QDict *options,
307 const char *fmt, int flags,
308 bool writethrough, bool quiet,
309 bool force_share)
310 {
311 BlockBackend *blk;
312 Error *local_err = NULL;
313
314 if (!options) {
315 options = qdict_new();
316 }
317 if (fmt) {
318 qdict_put_str(options, "driver", fmt);
319 }
320
321 if (force_share) {
322 qdict_put_bool(options, BDRV_OPT_FORCE_SHARE, true);
323 }
324 blk = blk_new_open(filename, NULL, options, flags, &local_err);
325 if (!blk) {
326 error_reportf_err(local_err, "Could not open '%s': ", filename);
327 return NULL;
328 }
329 blk_set_enable_write_cache(blk, !writethrough);
330
331 return blk;
332 }
333
334
335 static int img_add_key_secrets(void *opaque,
336 const char *name, const char *value,
337 Error **errp)
338 {
339 QDict *options = opaque;
340
341 if (g_str_has_suffix(name, "key-secret")) {
342 qdict_put_str(options, name, value);
343 }
344
345 return 0;
346 }
347
348
349 static BlockBackend *img_open(bool image_opts,
350 const char *filename,
351 const char *fmt, int flags, bool writethrough,
352 bool quiet, bool force_share)
353 {
354 BlockBackend *blk;
355 if (image_opts) {
356 QemuOpts *opts;
357 if (fmt) {
358 error_report("--image-opts and --format are mutually exclusive");
359 return NULL;
360 }
361 opts = qemu_opts_parse_noisily(qemu_find_opts("source"),
362 filename, true);
363 if (!opts) {
364 return NULL;
365 }
366 blk = img_open_opts(filename, opts, flags, writethrough, quiet,
367 force_share);
368 } else {
369 blk = img_open_file(filename, NULL, fmt, flags, writethrough, quiet,
370 force_share);
371 }
372
373 if (blk) {
374 blk_set_force_allow_inactivate(blk);
375 }
376
377 return blk;
378 }
379
380
381 static int add_old_style_options(const char *fmt, QemuOpts *opts,
382 const char *base_filename,
383 const char *base_fmt)
384 {
385 if (base_filename) {
386 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
387 NULL)) {
388 error_report("Backing file not supported for file format '%s'",
389 fmt);
390 return -1;
391 }
392 }
393 if (base_fmt) {
394 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
395 error_report("Backing file format not supported for file "
396 "format '%s'", fmt);
397 return -1;
398 }
399 }
400 return 0;
401 }
402
403 static int64_t cvtnum_full(const char *name, const char *value,
404 bool is_size, int64_t min, int64_t max)
405 {
406 int err;
407 uint64_t res;
408
409 err = is_size ? qemu_strtosz(value, NULL, &res) :
410 qemu_strtou64(value, NULL, 0, &res);
411 if (err < 0 && err != -ERANGE) {
412 error_report("Invalid %s specified: '%s'", name, value);
413 return err;
414 }
415 if (err == -ERANGE || res > max || res < min) {
416 error_report("Invalid %s specified. Must be between %" PRId64
417 " and %" PRId64 ".", name, min, max);
418 return -ERANGE;
419 }
420 return res;
421 }
422
423 static int64_t cvtnum(const char *name, const char *value, bool is_size)
424 {
425 return cvtnum_full(name, value, is_size, 0, INT64_MAX);
426 }
427
428 static int img_create(const img_cmd_t *ccmd, int argc, char **argv)
429 {
430 int c;
431 int64_t img_size = -1;
432 const char *fmt = "raw";
433 const char *base_fmt = NULL;
434 const char *filename;
435 const char *base_filename = NULL;
436 char *options = NULL;
437 Error *local_err = NULL;
438 bool quiet = false;
439 int flags = 0;
440
441 for(;;) {
442 static const struct option long_options[] = {
443 {"help", no_argument, 0, 'h'},
444 {"format", required_argument, 0, 'f'},
445 {"options", required_argument, 0, 'o'},
446 {"backing", required_argument, 0, 'b'},
447 {"backing-format", required_argument, 0, 'B'}, /* was -F in 10.0 */
448 {"backing-unsafe", no_argument, 0, 'u'},
449 {"quiet", no_argument, 0, 'q'},
450 {"object", required_argument, 0, OPTION_OBJECT},
451 {0, 0, 0, 0}
452 };
453 c = getopt_long(argc, argv, "hf:o:b:F:B:uq",
454 long_options, NULL);
455 if (c == -1) {
456 break;
457 }
458 switch(c) {
459 case 'h':
460 cmd_help(ccmd, "[-f FMT] [-o FMT_OPTS]\n"
461 " [-b BACKING_FILE [-B BACKING_FMT]] [-u]\n"
462 " [-q] [--object OBJDEF] FILE [SIZE]\n"
463 ,
464 " -f, --format FMT\n"
465 " specifies the format of the new image (default: raw)\n"
466 " -o, --options FMT_OPTS\n"
467 " format-specific options (specify '-o help' for help)\n"
468 " -b, --backing BACKING_FILE\n"
469 " create target image to be a CoW on top of BACKING_FILE\n"
470 " -B, --backing-format BACKING_FMT (was -F in <= 10.0)\n"
471 " specifies the format of BACKING_FILE (default: probing is used)\n"
472 " -u, --backing-unsafe\n"
473 " do not fail if BACKING_FILE can not be read\n"
474 " -q, --quiet\n"
475 " quiet mode (produce only error messages if any)\n"
476 " --object OBJDEF\n"
477 " defines QEMU user-creatable object\n"
478 " FILE\n"
479 " name of the image file to create (will be overritten if already exists)\n"
480 " SIZE[bKMGTPE]\n"
481 " image size with optional multiplier suffix (powers of 1024)\n"
482 " (required unless BACKING_FILE is specified)\n"
483 );
484 break;
485 case 'f':
486 fmt = optarg;
487 break;
488 case 'o':
489 if (accumulate_options(&options, optarg) < 0) {
490 goto fail;
491 }
492 break;
493 case 'b':
494 base_filename = optarg;
495 break;
496 case 'F': /* <=10.0 */
497 case 'B':
498 base_fmt = optarg;
499 break;
500 case 'u':
501 flags |= BDRV_O_NO_BACKING;
502 break;
503 case 'q':
504 quiet = true;
505 break;
506 case OPTION_OBJECT:
507 user_creatable_process_cmdline(optarg);
508 break;
509 default:
510 tryhelp(argv[0]);
511 }
512 }
513
514 /* Get the filename */
515 filename = (optind < argc) ? argv[optind] : NULL;
516 if (options && has_help_option(options)) {
517 g_free(options);
518 return print_block_option_help(filename, fmt);
519 }
520
521 if (optind >= argc) {
522 error_exit(argv[0], "Expecting image file name");
523 }
524 optind++;
525
526 /* Get image size, if specified */
527 if (optind < argc) {
528 img_size = cvtnum("image size", argv[optind++], true);
529 if (img_size < 0) {
530 goto fail;
531 }
532 }
533 if (optind != argc) {
534 error_exit(argv[0], "Unexpected argument: %s", argv[optind]);
535 }
536
537 bdrv_img_create(filename, fmt, base_filename, base_fmt,
538 options, img_size, flags, quiet, &local_err);
539 if (local_err) {
540 error_reportf_err(local_err, "%s: ", filename);
541 goto fail;
542 }
543
544 g_free(options);
545 return 0;
546
547 fail:
548 g_free(options);
549 return 1;
550 }
551
552 static void dump_json_image_check(ImageCheck *check, bool quiet)
553 {
554 GString *str;
555 QObject *obj;
556 Visitor *v = qobject_output_visitor_new(&obj);
557
558 visit_type_ImageCheck(v, NULL, &check, &error_abort);
559 visit_complete(v, &obj);
560 str = qobject_to_json_pretty(obj, true);
561 assert(str != NULL);
562 qprintf(quiet, "%s\n", str->str);
563 qobject_unref(obj);
564 visit_free(v);
565 g_string_free(str, true);
566 }
567
568 static void dump_human_image_check(ImageCheck *check, bool quiet)
569 {
570 if (!(check->corruptions || check->leaks || check->check_errors)) {
571 qprintf(quiet, "No errors were found on the image.\n");
572 } else {
573 if (check->corruptions) {
574 qprintf(quiet, "\n%" PRId64 " errors were found on the image.\n"
575 "Data may be corrupted, or further writes to the image "
576 "may corrupt it.\n",
577 check->corruptions);
578 }
579
580 if (check->leaks) {
581 qprintf(quiet,
582 "\n%" PRId64 " leaked clusters were found on the image.\n"
583 "This means waste of disk space, but no harm to data.\n",
584 check->leaks);
585 }
586
587 if (check->check_errors) {
588 qprintf(quiet,
589 "\n%" PRId64
590 " internal errors have occurred during the check.\n",
591 check->check_errors);
592 }
593 }
594
595 if (check->total_clusters != 0 && check->allocated_clusters != 0) {
596 qprintf(quiet, "%" PRId64 "/%" PRId64 " = %0.2f%% allocated, "
597 "%0.2f%% fragmented, %0.2f%% compressed clusters\n",
598 check->allocated_clusters, check->total_clusters,
599 check->allocated_clusters * 100.0 / check->total_clusters,
600 check->fragmented_clusters * 100.0 / check->allocated_clusters,
601 check->compressed_clusters * 100.0 /
602 check->allocated_clusters);
603 }
604
605 if (check->image_end_offset) {
606 qprintf(quiet,
607 "Image end offset: %" PRId64 "\n", check->image_end_offset);
608 }
609 }
610
611 static int collect_image_check(BlockDriverState *bs,
612 ImageCheck *check,
613 const char *filename,
614 const char *fmt,
615 int fix)
616 {
617 int ret;
618 BdrvCheckResult result;
619
620 ret = bdrv_check(bs, &result, fix);
621 if (ret < 0) {
622 return ret;
623 }
624
625 check->filename = g_strdup(filename);
626 check->format = g_strdup(bdrv_get_format_name(bs));
627 check->check_errors = result.check_errors;
628 check->corruptions = result.corruptions;
629 check->has_corruptions = result.corruptions != 0;
630 check->leaks = result.leaks;
631 check->has_leaks = result.leaks != 0;
632 check->corruptions_fixed = result.corruptions_fixed;
633 check->has_corruptions_fixed = result.corruptions_fixed != 0;
634 check->leaks_fixed = result.leaks_fixed;
635 check->has_leaks_fixed = result.leaks_fixed != 0;
636 check->image_end_offset = result.image_end_offset;
637 check->has_image_end_offset = result.image_end_offset != 0;
638 check->total_clusters = result.bfi.total_clusters;
639 check->has_total_clusters = result.bfi.total_clusters != 0;
640 check->allocated_clusters = result.bfi.allocated_clusters;
641 check->has_allocated_clusters = result.bfi.allocated_clusters != 0;
642 check->fragmented_clusters = result.bfi.fragmented_clusters;
643 check->has_fragmented_clusters = result.bfi.fragmented_clusters != 0;
644 check->compressed_clusters = result.bfi.compressed_clusters;
645 check->has_compressed_clusters = result.bfi.compressed_clusters != 0;
646
647 return 0;
648 }
649
650 /*
651 * Checks an image for consistency. Exit codes:
652 *
653 * 0 - Check completed, image is good
654 * 1 - Check not completed because of internal errors
655 * 2 - Check completed, image is corrupted
656 * 3 - Check completed, image has leaked clusters, but is good otherwise
657 * 63 - Checks are not supported by the image format
658 */
659 static int img_check(const img_cmd_t *ccmd, int argc, char **argv)
660 {
661 int c, ret;
662 OutputFormat output_format = OFORMAT_HUMAN;
663 const char *filename, *fmt, *cache;
664 BlockBackend *blk;
665 BlockDriverState *bs;
666 int fix = 0;
667 int flags = BDRV_O_CHECK;
668 bool writethrough;
669 ImageCheck *check;
670 bool quiet = false;
671 bool image_opts = false;
672 bool force_share = false;
673
674 fmt = NULL;
675 cache = BDRV_DEFAULT_CACHE;
676
677 for(;;) {
678 int option_index = 0;
679 static const struct option long_options[] = {
680 {"help", no_argument, 0, 'h'},
681 {"format", required_argument, 0, 'f'},
682 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
683 {"cache", required_argument, 0, 'T'},
684 {"repair", required_argument, 0, 'r'},
685 {"force-share", no_argument, 0, 'U'},
686 {"output", required_argument, 0, OPTION_OUTPUT},
687 {"quiet", no_argument, 0, 'q'},
688 {"object", required_argument, 0, OPTION_OBJECT},
689 {0, 0, 0, 0}
690 };
691 c = getopt_long(argc, argv, "hf:T:r:Uq",
692 long_options, &option_index);
693 if (c == -1) {
694 break;
695 }
696 switch(c) {
697 case 'h':
698 cmd_help(ccmd, "[-f FMT | --image-opts] [-T CACHE_MODE] [-r leaks|all]\n"
699 " [-U] [--output human|json] [-q] [--object OBJDEF] FILE\n"
700 ,
701 " -f, --format FMT\n"
702 " specifies the format of the image explicitly (default: probing is used)\n"
703 " --image-opts\n"
704 " treat FILE as an option string (key=value,..), not a file name\n"
705 " (incompatible with -f|--format)\n"
706 " -T, --cache CACHE_MODE\n" /* why not -t ? */
707 " cache mode (default: " BDRV_DEFAULT_CACHE ")\n"
708 " -r, --repair leaks|all\n"
709 " repair errors of the given category in the image (image will be\n"
710 " opened in read-write mode, incompatible with -U|--force-share)\n"
711 " -U, --force-share\n"
712 " open image in shared mode for concurrent access\n"
713 " --output human|json\n"
714 " output format (default: human)\n"
715 " -q, --quiet\n"
716 " quiet mode (produce only error messages if any)\n"
717 " --object OBJDEF\n"
718 " defines QEMU user-creatable object\n"
719 " FILE\n"
720 " name of the image file, or an option string (key=value,..)\n"
721 " with --image-opts, to operate on\n"
722 );
723 break;
724 case 'f':
725 fmt = optarg;
726 break;
727 case OPTION_IMAGE_OPTS:
728 image_opts = true;
729 break;
730 case 'T':
731 cache = optarg;
732 break;
733 case 'r':
734 flags |= BDRV_O_RDWR;
735
736 if (!strcmp(optarg, "leaks")) {
737 fix = BDRV_FIX_LEAKS;
738 } else if (!strcmp(optarg, "all")) {
739 fix = BDRV_FIX_LEAKS | BDRV_FIX_ERRORS;
740 } else {
741 error_exit(argv[0],
742 "--repair (-r) expects 'leaks' or 'all', not '%s'",
743 optarg);
744 }
745 break;
746 case 'U':
747 force_share = true;
748 break;
749 case OPTION_OUTPUT:
750 output_format = parse_output_format(argv[0], optarg);
751 break;
752 case 'q':
753 quiet = true;
754 break;
755 case OPTION_OBJECT:
756 user_creatable_process_cmdline(optarg);
757 break;
758 default:
759 tryhelp(argv[0]);
760 }
761 }
762 if (optind != argc - 1) {
763 error_exit(argv[0], "Expecting one image file name");
764 }
765 filename = argv[optind++];
766
767 ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
768 if (ret < 0) {
769 error_report("Invalid source cache option: %s", cache);
770 return 1;
771 }
772
773 blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
774 force_share);
775 if (!blk) {
776 return 1;
777 }
778 bs = blk_bs(blk);
779
780 check = g_new0(ImageCheck, 1);
781 ret = collect_image_check(bs, check, filename, fmt, fix);
782
783 if (ret == -ENOTSUP) {
784 error_report("This image format does not support checks");
785 ret = 63;
786 goto fail;
787 }
788
789 if (check->corruptions_fixed || check->leaks_fixed) {
790 int corruptions_fixed, leaks_fixed;
791 bool has_leaks_fixed, has_corruptions_fixed;
792
793 leaks_fixed = check->leaks_fixed;
794 has_leaks_fixed = check->has_leaks_fixed;
795 corruptions_fixed = check->corruptions_fixed;
796 has_corruptions_fixed = check->has_corruptions_fixed;
797
798 if (output_format == OFORMAT_HUMAN) {
799 qprintf(quiet,
800 "The following inconsistencies were found and repaired:\n\n"
801 " %" PRId64 " leaked clusters\n"
802 " %" PRId64 " corruptions\n\n"
803 "Double checking the fixed image now...\n",
804 check->leaks_fixed,
805 check->corruptions_fixed);
806 }
807
808 qapi_free_ImageCheck(check);
809 check = g_new0(ImageCheck, 1);
810 ret = collect_image_check(bs, check, filename, fmt, 0);
811
812 check->leaks_fixed = leaks_fixed;
813 check->has_leaks_fixed = has_leaks_fixed;
814 check->corruptions_fixed = corruptions_fixed;
815 check->has_corruptions_fixed = has_corruptions_fixed;
816 }
817
818 if (!ret) {
819 switch (output_format) {
820 case OFORMAT_HUMAN:
821 dump_human_image_check(check, quiet);
822 break;
823 case OFORMAT_JSON:
824 dump_json_image_check(check, quiet);
825 break;
826 }
827 }
828
829 if (ret || check->check_errors) {
830 if (ret) {
831 error_report("Check failed: %s", strerror(-ret));
832 } else {
833 error_report("Check failed");
834 }
835 ret = 1;
836 goto fail;
837 }
838
839 if (check->corruptions) {
840 ret = 2;
841 } else if (check->leaks) {
842 ret = 3;
843 } else {
844 ret = 0;
845 }
846
847 fail:
848 qapi_free_ImageCheck(check);
849 blk_unref(blk);
850 return ret;
851 }
852
853 typedef struct CommonBlockJobCBInfo {
854 BlockDriverState *bs;
855 Error **errp;
856 } CommonBlockJobCBInfo;
857
858 static void common_block_job_cb(void *opaque, int ret)
859 {
860 CommonBlockJobCBInfo *cbi = opaque;
861
862 if (ret < 0) {
863 error_setg_errno(cbi->errp, -ret, "Block job failed");
864 }
865 }
866
867 static void run_block_job(BlockJob *job, Error **errp)
868 {
869 uint64_t progress_current, progress_total;
870 AioContext *aio_context = block_job_get_aio_context(job);
871 int ret = 0;
872
873 job_lock();
874 job_ref_locked(&job->job);
875 do {
876 float progress = 0.0f;
877 job_unlock();
878 aio_poll(aio_context, true);
879
880 progress_get_snapshot(&job->job.progress, &progress_current,
881 &progress_total);
882 if (progress_total) {
883 progress = (float)progress_current / progress_total * 100.f;
884 }
885 qemu_progress_print(progress, 0);
886 job_lock();
887 } while (!job_is_ready_locked(&job->job) &&
888 !job_is_completed_locked(&job->job));
889
890 if (!job_is_completed_locked(&job->job)) {
891 ret = job_complete_sync_locked(&job->job, errp);
892 } else {
893 ret = job->job.ret;
894 }
895 job_unref_locked(&job->job);
896 job_unlock();
897
898 /* publish completion progress only when success */
899 if (!ret) {
900 qemu_progress_print(100.f, 0);
901 }
902 }
903
904 static int img_commit(const img_cmd_t *ccmd, int argc, char **argv)
905 {
906 int c, ret, flags;
907 const char *filename, *fmt, *cache, *base;
908 BlockBackend *blk;
909 BlockDriverState *bs, *base_bs;
910 BlockJob *job;
911 bool progress = false, quiet = false, drop = false;
912 bool writethrough;
913 Error *local_err = NULL;
914 CommonBlockJobCBInfo cbi;
915 bool image_opts = false;
916 int64_t rate_limit = 0;
917
918 fmt = NULL;
919 cache = BDRV_DEFAULT_CACHE;
920 base = NULL;
921 for(;;) {
922 static const struct option long_options[] = {
923 {"help", no_argument, 0, 'h'},
924 {"format", required_argument, 0, 'f'},
925 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
926 {"cache", required_argument, 0, 't'},
927 {"drop", no_argument, 0, 'd'},
928 {"base", required_argument, 0, 'b'},
929 {"rate-limit", required_argument, 0, 'r'},
930 {"progress", no_argument, 0, 'p'},
931 {"quiet", no_argument, 0, 'q'},
932 {"object", required_argument, 0, OPTION_OBJECT},
933 {0, 0, 0, 0}
934 };
935 c = getopt_long(argc, argv, "hf:t:db:r:pq",
936 long_options, NULL);
937 if (c == -1) {
938 break;
939 }
940 switch(c) {
941 case 'h':
942 cmd_help(ccmd, "[-f FMT | --image-opts] [-t CACHE_MODE] [-b BASE_IMG]\n"
943 " [-d] [-r RATE] [-q] [--object OBJDEF] FILE\n"
944 ,
945 " -f, --format FMT\n"
946 " specify FILE image format explicitly (default: probing is used)\n"
947 " --image-opts\n"
948 " treat FILE as an option string (key=value,..), not a file name\n"
949 " (incompatible with -f|--format)\n"
950 " -t, --cache CACHE_MODE image cache mode (default: " BDRV_DEFAULT_CACHE ")\n"
951 " -d, --drop\n"
952 " skip emptying FILE on completion\n"
953 " -b, --base BASE_IMG\n"
954 " image in the backing chain to commit change to\n"
955 " (default: immediate backing file; implies --drop)\n"
956 " -r, --rate-limit RATE\n"
957 " I/O rate limit, in bytes per second\n"
958 " -p, --progress\n"
959 " display progress information\n"
960 " -q, --quiet\n"
961 " quiet mode (produce only error messages if any)\n"
962 " --object OBJDEF\n"
963 " defines QEMU user-creatable object\n"
964 " FILE\n"
965 " name of the image file, or an option string (key=value,..)\n"
966 " with --image-opts, to operate on\n"
967 );
968 break;
969 case 'f':
970 fmt = optarg;
971 break;
972 case OPTION_IMAGE_OPTS:
973 image_opts = true;
974 break;
975 case 't':
976 cache = optarg;
977 break;
978 case 'd':
979 drop = true;
980 break;
981 case 'b':
982 base = optarg;
983 /* -b implies -d */
984 drop = true;
985 break;
986 case 'r':
987 rate_limit = cvtnum("rate limit", optarg, true);
988 if (rate_limit < 0) {
989 return 1;
990 }
991 break;
992 case 'p':
993 progress = true;
994 break;
995 case 'q':
996 quiet = true;
997 break;
998 case OPTION_OBJECT:
999 user_creatable_process_cmdline(optarg);
1000 break;
1001 default:
1002 tryhelp(argv[0]);
1003 }
1004 }
1005
1006 /* Progress is not shown in Quiet mode */
1007 if (quiet) {
1008 progress = false;
1009 }
1010
1011 if (optind != argc - 1) {
1012 error_exit(argv[0], "Expecting one image file name");
1013 }
1014 filename = argv[optind++];
1015
1016 flags = BDRV_O_RDWR | BDRV_O_UNMAP;
1017 ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
1018 if (ret < 0) {
1019 error_report("Invalid cache option: %s", cache);
1020 return 1;
1021 }
1022
1023 blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
1024 false);
1025 if (!blk) {
1026 return 1;
1027 }
1028 bs = blk_bs(blk);
1029
1030 qemu_progress_init(progress, 1.f);
1031 qemu_progress_print(0.f, 100);
1032
1033 bdrv_graph_rdlock_main_loop();
1034 if (base) {
1035 base_bs = bdrv_find_backing_image(bs, base);
1036 if (!base_bs) {
1037 error_setg(&local_err,
1038 "Did not find '%s' in the backing chain of '%s'",
1039 base, filename);
1040 bdrv_graph_rdunlock_main_loop();
1041 goto done;
1042 }
1043 } else {
1044 /* This is different from QMP, which by default uses the deepest file in
1045 * the backing chain (i.e., the very base); however, the traditional
1046 * behavior of qemu-img commit is using the immediate backing file. */
1047 base_bs = bdrv_backing_chain_next(bs);
1048 if (!base_bs) {
1049 error_setg(&local_err, "Image does not have a backing file");
1050 bdrv_graph_rdunlock_main_loop();
1051 goto done;
1052 }
1053 }
1054 bdrv_graph_rdunlock_main_loop();
1055
1056 cbi = (CommonBlockJobCBInfo){
1057 .errp = &local_err,
1058 .bs = bs,
1059 };
1060
1061 commit_active_start("commit", bs, base_bs, JOB_DEFAULT, rate_limit,
1062 BLOCKDEV_ON_ERROR_REPORT, NULL, common_block_job_cb,
1063 &cbi, false, &local_err);
1064 if (local_err) {
1065 goto done;
1066 }
1067
1068 /* When the block job completes, the BlockBackend reference will point to
1069 * the old backing file. In order to avoid that the top image is already
1070 * deleted, so we can still empty it afterwards, increment the reference
1071 * counter here preemptively. */
1072 if (!drop) {
1073 bdrv_ref(bs);
1074 }
1075
1076 job = block_job_get("commit");
1077 assert(job);
1078 run_block_job(job, &local_err);
1079 if (local_err) {
1080 goto unref_backing;
1081 }
1082
1083 if (!drop) {
1084 BlockBackend *old_backing_blk;
1085
1086 old_backing_blk = blk_new_with_bs(bs, BLK_PERM_WRITE, BLK_PERM_ALL,
1087 &local_err);
1088 if (!old_backing_blk) {
1089 goto unref_backing;
1090 }
1091 ret = blk_make_empty(old_backing_blk, &local_err);
1092 blk_unref(old_backing_blk);
1093 if (ret == -ENOTSUP) {
1094 error_free(local_err);
1095 local_err = NULL;
1096 } else if (ret < 0) {
1097 goto unref_backing;
1098 }
1099 }
1100
1101 unref_backing:
1102 if (!drop) {
1103 bdrv_unref(bs);
1104 }
1105
1106 done:
1107 qemu_progress_end();
1108
1109 /*
1110 * Manually inactivate the image first because this way we can know whether
1111 * an error occurred. blk_unref() doesn't tell us about failures.
1112 */
1113 ret = bdrv_inactivate_all();
1114 if (ret < 0 && !local_err) {
1115 error_setg_errno(&local_err, -ret, "Error while closing the image");
1116 }
1117 blk_unref(blk);
1118
1119 if (local_err) {
1120 error_report_err(local_err);
1121 return 1;
1122 }
1123
1124 qprintf(quiet, "Image committed.\n");
1125 return 0;
1126 }
1127
1128 /*
1129 * Returns -1 if 'buf' contains only zeroes, otherwise the byte index
1130 * of the first sector boundary within buf where the sector contains a
1131 * non-zero byte. This function is robust to a buffer that is not
1132 * sector-aligned.
1133 */
1134 static int64_t find_nonzero(const uint8_t *buf, int64_t n)
1135 {
1136 int64_t i;
1137 int64_t end = QEMU_ALIGN_DOWN(n, BDRV_SECTOR_SIZE);
1138
1139 for (i = 0; i < end; i += BDRV_SECTOR_SIZE) {
1140 if (!buffer_is_zero(buf + i, BDRV_SECTOR_SIZE)) {
1141 return i;
1142 }
1143 }
1144 if (i < n && !buffer_is_zero(buf + i, n - end)) {
1145 return i;
1146 }
1147 return -1;
1148 }
1149
1150 /*
1151 * Returns true iff the first sector pointed to by 'buf' contains at least
1152 * a non-NUL byte.
1153 *
1154 * 'pnum' is set to the number of sectors (including and immediately following
1155 * the first one) that are known to be in the same allocated/unallocated state.
1156 * The function will try to align the end offset to alignment boundaries so
1157 * that the request will at least end aligned and consecutive requests will
1158 * also start at an aligned offset.
1159 */
1160 static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum,
1161 int64_t sector_num, int alignment)
1162 {
1163 bool is_zero;
1164 int i, tail;
1165
1166 if (n <= 0) {
1167 *pnum = 0;
1168 return 0;
1169 }
1170 is_zero = buffer_is_zero(buf, BDRV_SECTOR_SIZE);
1171 for(i = 1; i < n; i++) {
1172 buf += BDRV_SECTOR_SIZE;
1173 if (is_zero != buffer_is_zero(buf, BDRV_SECTOR_SIZE)) {
1174 break;
1175 }
1176 }
1177
1178 if (i == n) {
1179 /*
1180 * The whole buf is the same.
1181 * No reason to split it into chunks, so return now.
1182 */
1183 *pnum = i;
1184 return !is_zero;
1185 }
1186
1187 tail = (sector_num + i) & (alignment - 1);
1188 if (tail) {
1189 if (is_zero && i <= tail) {
1190 /*
1191 * For sure next sector after i is data, and it will rewrite this
1192 * tail anyway due to RMW. So, let's just write data now.
1193 */
1194 is_zero = false;
1195 }
1196 if (!is_zero) {
1197 /* If possible, align up end offset of allocated areas. */
1198 i += alignment - tail;
1199 i = MIN(i, n);
1200 } else {
1201 /*
1202 * For sure next sector after i is data, and it will rewrite this
1203 * tail anyway due to RMW. Better is avoid RMW and write zeroes up
1204 * to aligned bound.
1205 */
1206 i -= tail;
1207 }
1208 }
1209 *pnum = i;
1210 return !is_zero;
1211 }
1212
1213 /*
1214 * Like is_allocated_sectors, but if the buffer starts with a used sector,
1215 * up to 'min' consecutive sectors containing zeros are ignored. This avoids
1216 * breaking up write requests for only small sparse areas.
1217 */
1218 static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum,
1219 int min, int64_t sector_num, int alignment)
1220 {
1221 int ret;
1222 int num_checked, num_used;
1223
1224 if (n < min) {
1225 min = n;
1226 }
1227
1228 ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
1229 if (!ret) {
1230 return ret;
1231 }
1232
1233 num_used = *pnum;
1234 buf += BDRV_SECTOR_SIZE * *pnum;
1235 n -= *pnum;
1236 sector_num += *pnum;
1237 num_checked = num_used;
1238
1239 while (n > 0) {
1240 ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
1241
1242 buf += BDRV_SECTOR_SIZE * *pnum;
1243 n -= *pnum;
1244 sector_num += *pnum;
1245 num_checked += *pnum;
1246 if (ret) {
1247 num_used = num_checked;
1248 } else if (*pnum >= min) {
1249 break;
1250 }
1251 }
1252
1253 *pnum = num_used;
1254 return 1;
1255 }
1256
1257 /*
1258 * Compares two buffers chunk by chunk, where @chsize is the chunk size.
1259 * If @chsize is 0, default chunk size of BDRV_SECTOR_SIZE is used.
1260 * Returns 0 if the first chunk of each buffer matches, non-zero otherwise.
1261 *
1262 * @pnum is set to the size of the buffer prefix aligned to @chsize that
1263 * has the same matching status as the first chunk.
1264 */
1265 static int compare_buffers(const uint8_t *buf1, const uint8_t *buf2,
1266 int64_t bytes, uint64_t chsize, int64_t *pnum)
1267 {
1268 bool res;
1269 int64_t i;
1270
1271 assert(bytes > 0);
1272
1273 if (!chsize) {
1274 chsize = BDRV_SECTOR_SIZE;
1275 }
1276 i = MIN(bytes, chsize);
1277
1278 res = !!memcmp(buf1, buf2, i);
1279 while (i < bytes) {
1280 int64_t len = MIN(bytes - i, chsize);
1281
1282 if (!!memcmp(buf1 + i, buf2 + i, len) != res) {
1283 break;
1284 }
1285 i += len;
1286 }
1287
1288 *pnum = i;
1289 return res;
1290 }
1291
1292 #define IO_BUF_SIZE (2 * MiB)
1293
1294 /*
1295 * Check if passed sectors are empty (not allocated or contain only 0 bytes)
1296 *
1297 * Intended for use by 'qemu-img compare': Returns 0 in case sectors are
1298 * filled with 0, 1 if sectors contain non-zero data (this is a comparison
1299 * failure), and 4 on error (the exit status for read errors), after emitting
1300 * an error message.
1301 *
1302 * @param blk: BlockBackend for the image
1303 * @param offset: Starting offset to check
1304 * @param bytes: Number of bytes to check
1305 * @param filename: Name of disk file we are checking (logging purpose)
1306 * @param buffer: Allocated buffer for storing read data
1307 * @param quiet: Flag for quiet mode
1308 */
1309 static int check_empty_sectors(BlockBackend *blk, int64_t offset,
1310 int64_t bytes, const char *filename,
1311 uint8_t *buffer, bool quiet)
1312 {
1313 int ret = 0;
1314 int64_t idx;
1315
1316 ret = blk_pread(blk, offset, bytes, buffer, 0);
1317 if (ret < 0) {
1318 error_report("Error while reading offset %" PRId64 " of %s: %s",
1319 offset, filename, strerror(-ret));
1320 return 4;
1321 }
1322 idx = find_nonzero(buffer, bytes);
1323 if (idx >= 0) {
1324 qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1325 offset + idx);
1326 return 1;
1327 }
1328
1329 return 0;
1330 }
1331
1332 /*
1333 * Compares two images. Exit codes:
1334 *
1335 * 0 - Images are identical or the requested help was printed
1336 * 1 - Images differ
1337 * >1 - Error occurred
1338 */
1339 static int img_compare(const img_cmd_t *ccmd, int argc, char **argv)
1340 {
1341 const char *fmt1 = NULL, *fmt2 = NULL, *cache, *filename1, *filename2;
1342 BlockBackend *blk1, *blk2;
1343 BlockDriverState *bs1, *bs2;
1344 int64_t total_size1, total_size2;
1345 uint8_t *buf1 = NULL, *buf2 = NULL;
1346 int64_t pnum1, pnum2;
1347 int allocated1, allocated2;
1348 int ret = 0; /* return value - 0 Ident, 1 Different, >1 Error */
1349 bool progress = false, quiet = false, strict = false;
1350 int flags;
1351 bool writethrough;
1352 int64_t total_size;
1353 int64_t offset = 0;
1354 int64_t chunk;
1355 int c;
1356 uint64_t progress_base;
1357 bool image_opts = false;
1358 bool force_share = false;
1359
1360 cache = BDRV_DEFAULT_CACHE;
1361 for (;;) {
1362 static const struct option long_options[] = {
1363 {"help", no_argument, 0, 'h'},
1364 {"a-format", required_argument, 0, 'f'},
1365 {"b-format", required_argument, 0, 'F'},
1366 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
1367 {"strict", no_argument, 0, 's'},
1368 {"cache", required_argument, 0, 'T'},
1369 {"force-share", no_argument, 0, 'U'},
1370 {"progress", no_argument, 0, 'p'},
1371 {"quiet", no_argument, 0, 'q'},
1372 {"object", required_argument, 0, OPTION_OBJECT},
1373 {0, 0, 0, 0}
1374 };
1375 c = getopt_long(argc, argv, "hf:F:sT:Upq",
1376 long_options, NULL);
1377 if (c == -1) {
1378 break;
1379 }
1380 switch (c) {
1381 case 'h':
1382 cmd_help(ccmd,
1383 "[[-f FMT] [-F FMT] | --image-opts] [-s] [-T CACHE]\n"
1384 " [-U] [-p] [-q] [--object OBJDEF] FILE1 FILE2\n"
1385 ,
1386 " -f, --a-format FMT\n"
1387 " specify FILE1 image format explicitly (default: probing is used)\n"
1388 " -F, --b-format FMT\n"
1389 " specify FILE2 image format explicitly (default: probing is used)\n"
1390 " --image-opts\n"
1391 " treat FILE1 and FILE2 as option strings (key=value,..), not file names\n"
1392 " (incompatible with -f|--a-format and -F|--b-format)\n"
1393 " -s, --strict\n"
1394 " strict mode, also check if sizes are equal\n"
1395 " -T, --cache CACHE_MODE\n"
1396 " images caching mode (default: " BDRV_DEFAULT_CACHE ")\n"
1397 " -U, --force-share\n"
1398 " open images in shared mode for concurrent access\n"
1399 " -p, --progress\n"
1400 " display progress information\n"
1401 " -q, --quiet\n"
1402 " quiet mode (produce only error messages if any)\n"
1403 " --object OBJDEF\n"
1404 " defines QEMU user-creatable object\n"
1405 " FILE1, FILE2\n"
1406 " names of the image files, or option strings (key=value,..)\n"
1407 " with --image-opts, to compare\n"
1408 );
1409 break;
1410 case 'f':
1411 fmt1 = optarg;
1412 break;
1413 case 'F':
1414 fmt2 = optarg;
1415 break;
1416 case OPTION_IMAGE_OPTS:
1417 image_opts = true;
1418 break;
1419 case 's':
1420 strict = true;
1421 break;
1422 case 'T':
1423 cache = optarg;
1424 break;
1425 case 'U':
1426 force_share = true;
1427 break;
1428 case 'p':
1429 progress = true;
1430 break;
1431 case 'q':
1432 quiet = true;
1433 break;
1434 case OPTION_OBJECT:
1435 user_creatable_process_cmdline(optarg);
1436 break;
1437 default:
1438 tryhelp(argv[0]);
1439 }
1440 }
1441
1442 /* Progress is not shown in Quiet mode */
1443 if (quiet) {
1444 progress = false;
1445 }
1446
1447
1448 if (optind != argc - 2) {
1449 error_exit(argv[0], "Expecting two image file names");
1450 }
1451 filename1 = argv[optind++];
1452 filename2 = argv[optind++];
1453
1454 /* Initialize before goto out */
1455 qemu_progress_init(progress, 2.0);
1456
1457 flags = 0;
1458 ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
1459 if (ret < 0) {
1460 error_report("Invalid source cache option: %s", cache);
1461 ret = 2;
1462 goto out3;
1463 }
1464
1465 blk1 = img_open(image_opts, filename1, fmt1, flags, writethrough, quiet,
1466 force_share);
1467 if (!blk1) {
1468 ret = 2;
1469 goto out3;
1470 }
1471
1472 blk2 = img_open(image_opts, filename2, fmt2, flags, writethrough, quiet,
1473 force_share);
1474 if (!blk2) {
1475 ret = 2;
1476 goto out2;
1477 }
1478 bs1 = blk_bs(blk1);
1479 bs2 = blk_bs(blk2);
1480
1481 buf1 = blk_blockalign(blk1, IO_BUF_SIZE);
1482 buf2 = blk_blockalign(blk2, IO_BUF_SIZE);
1483 total_size1 = blk_getlength(blk1);
1484 if (total_size1 < 0) {
1485 error_report("Can't get size of %s: %s",
1486 filename1, strerror(-total_size1));
1487 ret = 4;
1488 goto out;
1489 }
1490 total_size2 = blk_getlength(blk2);
1491 if (total_size2 < 0) {
1492 error_report("Can't get size of %s: %s",
1493 filename2, strerror(-total_size2));
1494 ret = 4;
1495 goto out;
1496 }
1497 total_size = MIN(total_size1, total_size2);
1498 progress_base = MAX(total_size1, total_size2);
1499
1500 qemu_progress_print(0, 100);
1501
1502 if (strict && total_size1 != total_size2) {
1503 ret = 1;
1504 qprintf(quiet, "Strict mode: Image size mismatch!\n");
1505 goto out;
1506 }
1507
1508 while (offset < total_size) {
1509 int status1, status2;
1510
1511 status1 = bdrv_block_status_above(bs1, NULL, offset,
1512 total_size1 - offset, &pnum1, NULL,
1513 NULL);
1514 if (status1 < 0) {
1515 ret = 3;
1516 error_report("Sector allocation test failed for %s", filename1);
1517 goto out;
1518 }
1519 allocated1 = status1 & BDRV_BLOCK_ALLOCATED;
1520
1521 status2 = bdrv_block_status_above(bs2, NULL, offset,
1522 total_size2 - offset, &pnum2, NULL,
1523 NULL);
1524 if (status2 < 0) {
1525 ret = 3;
1526 error_report("Sector allocation test failed for %s", filename2);
1527 goto out;
1528 }
1529 allocated2 = status2 & BDRV_BLOCK_ALLOCATED;
1530
1531 assert(pnum1 && pnum2);
1532 chunk = MIN(pnum1, pnum2);
1533
1534 if (strict) {
1535 if (status1 != status2) {
1536 ret = 1;
1537 qprintf(quiet, "Strict mode: Offset %" PRId64
1538 " block status mismatch!\n", offset);
1539 goto out;
1540 }
1541 }
1542 if ((status1 & BDRV_BLOCK_ZERO) && (status2 & BDRV_BLOCK_ZERO)) {
1543 /* nothing to do */
1544 } else if (allocated1 == allocated2) {
1545 if (allocated1) {
1546 int64_t pnum;
1547
1548 chunk = MIN(chunk, IO_BUF_SIZE);
1549 ret = blk_pread(blk1, offset, chunk, buf1, 0);
1550 if (ret < 0) {
1551 error_report("Error while reading offset %" PRId64
1552 " of %s: %s",
1553 offset, filename1, strerror(-ret));
1554 ret = 4;
1555 goto out;
1556 }
1557 ret = blk_pread(blk2, offset, chunk, buf2, 0);
1558 if (ret < 0) {
1559 error_report("Error while reading offset %" PRId64
1560 " of %s: %s",
1561 offset, filename2, strerror(-ret));
1562 ret = 4;
1563 goto out;
1564 }
1565 ret = compare_buffers(buf1, buf2, chunk, 0, &pnum);
1566 if (ret || pnum != chunk) {
1567 qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
1568 offset + (ret ? 0 : pnum));
1569 ret = 1;
1570 goto out;
1571 }
1572 }
1573 } else {
1574 chunk = MIN(chunk, IO_BUF_SIZE);
1575 if (allocated1) {
1576 ret = check_empty_sectors(blk1, offset, chunk,
1577 filename1, buf1, quiet);
1578 } else {
1579 ret = check_empty_sectors(blk2, offset, chunk,
1580 filename2, buf1, quiet);
1581 }
1582 if (ret) {
1583 goto out;
1584 }
1585 }
1586 offset += chunk;
1587 qemu_progress_print(((float) chunk / progress_base) * 100, 100);
1588 }
1589
1590 if (total_size1 != total_size2) {
1591 BlockBackend *blk_over;
1592 const char *filename_over;
1593
1594 qprintf(quiet, "Warning: Image size mismatch!\n");
1595 if (total_size1 > total_size2) {
1596 blk_over = blk1;
1597 filename_over = filename1;
1598 } else {
1599 blk_over = blk2;
1600 filename_over = filename2;
1601 }
1602
1603 while (offset < progress_base) {
1604 ret = bdrv_block_status_above(blk_bs(blk_over), NULL, offset,
1605 progress_base - offset, &chunk,
1606 NULL, NULL);
1607 if (ret < 0) {
1608 ret = 3;
1609 error_report("Sector allocation test failed for %s",
1610 filename_over);
1611 goto out;
1612
1613 }
1614 if (ret & BDRV_BLOCK_ALLOCATED && !(ret & BDRV_BLOCK_ZERO)) {
1615 chunk = MIN(chunk, IO_BUF_SIZE);
1616 ret = check_empty_sectors(blk_over, offset, chunk,
1617 filename_over, buf1, quiet);
1618 if (ret) {
1619 goto out;
1620 }
1621 }
1622 offset += chunk;
1623 qemu_progress_print(((float) chunk / progress_base) * 100, 100);
1624 }
1625 }
1626
1627 qprintf(quiet, "Images are identical.\n");
1628 ret = 0;
1629
1630 out:
1631 qemu_vfree(buf1);
1632 qemu_vfree(buf2);
1633 blk_unref(blk2);
1634 out2:
1635 blk_unref(blk1);
1636 out3:
1637 qemu_progress_end();
1638 return ret;
1639 }
1640
1641 /* Convenience wrapper around qmp_block_dirty_bitmap_merge */
1642 static void do_dirty_bitmap_merge(const char *dst_node, const char *dst_name,
1643 const char *src_node, const char *src_name,
1644 Error **errp)
1645 {
1646 BlockDirtyBitmapOrStr *merge_src;
1647 BlockDirtyBitmapOrStrList *list = NULL;
1648
1649 merge_src = g_new0(BlockDirtyBitmapOrStr, 1);
1650 merge_src->type = QTYPE_QDICT;
1651 merge_src->u.external.node = g_strdup(src_node);
1652 merge_src->u.external.name = g_strdup(src_name);
1653 QAPI_LIST_PREPEND(list, merge_src);
1654 qmp_block_dirty_bitmap_merge(dst_node, dst_name, list, errp);
1655 qapi_free_BlockDirtyBitmapOrStrList(list);
1656 }
1657
1658 enum ImgConvertBlockStatus {
1659 BLK_DATA,
1660 BLK_ZERO,
1661 BLK_BACKING_FILE,
1662 };
1663
1664 #define MAX_COROUTINES 16
1665 #define CONVERT_THROTTLE_GROUP "img_convert"
1666
1667 typedef struct ImgConvertState {
1668 BlockBackend **src;
1669 int64_t *src_sectors;
1670 int *src_alignment;
1671 int src_num;
1672 int64_t total_sectors;
1673 int64_t allocated_sectors;
1674 int64_t allocated_done;
1675 int64_t sector_num;
1676 int64_t wr_offs;
1677 enum ImgConvertBlockStatus status;
1678 int64_t sector_next_status;
1679 BlockBackend *target;
1680 bool has_zero_init;
1681 bool compressed;
1682 bool target_is_new;
1683 bool target_has_backing;
1684 int64_t target_backing_sectors; /* negative if unknown */
1685 bool wr_in_order;
1686 bool copy_range;
1687 bool salvage;
1688 bool quiet;
1689 int min_sparse;
1690 int alignment;
1691 size_t cluster_sectors;
1692 size_t buf_sectors;
1693 long num_coroutines;
1694 int running_coroutines;
1695 Coroutine *co[MAX_COROUTINES];
1696 int64_t wait_sector_num[MAX_COROUTINES];
1697 CoMutex lock;
1698 int ret;
1699 } ImgConvertState;
1700
1701 static void convert_select_part(ImgConvertState *s, int64_t sector_num,
1702 int *src_cur, int64_t *src_cur_offset)
1703 {
1704 *src_cur = 0;
1705 *src_cur_offset = 0;
1706 while (sector_num - *src_cur_offset >= s->src_sectors[*src_cur]) {
1707 *src_cur_offset += s->src_sectors[*src_cur];
1708 (*src_cur)++;
1709 assert(*src_cur < s->src_num);
1710 }
1711 }
1712
1713 static int coroutine_mixed_fn GRAPH_RDLOCK
1714 convert_iteration_sectors(ImgConvertState *s, int64_t sector_num)
1715 {
1716 int64_t src_cur_offset;
1717 int ret, n, src_cur;
1718 bool post_backing_zero = false;
1719
1720 convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1721
1722 assert(s->total_sectors > sector_num);
1723 n = MIN(s->total_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
1724
1725 if (s->target_backing_sectors >= 0) {
1726 if (sector_num >= s->target_backing_sectors) {
1727 post_backing_zero = true;
1728 } else if (sector_num + n > s->target_backing_sectors) {
1729 /* Split requests around target_backing_sectors (because
1730 * starting from there, zeros are handled differently) */
1731 n = s->target_backing_sectors - sector_num;
1732 }
1733 }
1734
1735 if (s->sector_next_status <= sector_num) {
1736 uint64_t offset = (sector_num - src_cur_offset) * BDRV_SECTOR_SIZE;
1737 int64_t count;
1738 int tail;
1739 BlockDriverState *src_bs = blk_bs(s->src[src_cur]);
1740 BlockDriverState *base;
1741
1742 if (s->target_has_backing) {
1743 base = bdrv_cow_bs(bdrv_skip_filters(src_bs));
1744 } else {
1745 base = NULL;
1746 }
1747
1748 do {
1749 count = n * BDRV_SECTOR_SIZE;
1750
1751 ret = bdrv_block_status_above(src_bs, base, offset, count, &count,
1752 NULL, NULL);
1753
1754 if (ret < 0) {
1755 if (s->salvage) {
1756 if (n == 1) {
1757 if (!s->quiet) {
1758 warn_report("error while reading block status at "
1759 "offset %" PRIu64 ": %s", offset,
1760 strerror(-ret));
1761 }
1762 /* Just try to read the data, then */
1763 ret = BDRV_BLOCK_DATA;
1764 count = BDRV_SECTOR_SIZE;
1765 } else {
1766 /* Retry on a shorter range */
1767 n = DIV_ROUND_UP(n, 4);
1768 }
1769 } else {
1770 error_report("error while reading block status at offset "
1771 "%" PRIu64 ": %s", offset, strerror(-ret));
1772 return ret;
1773 }
1774 }
1775 } while (ret < 0);
1776
1777 n = DIV_ROUND_UP(count, BDRV_SECTOR_SIZE);
1778
1779 /*
1780 * Avoid that s->sector_next_status becomes unaligned to the source
1781 * request alignment and/or cluster size to avoid unnecessary read
1782 * cycles.
1783 */
1784 tail = (sector_num - src_cur_offset + n) % s->src_alignment[src_cur];
1785 if (n > tail) {
1786 n -= tail;
1787 }
1788
1789 if (ret & BDRV_BLOCK_ZERO) {
1790 s->status = post_backing_zero ? BLK_BACKING_FILE : BLK_ZERO;
1791 } else if (ret & BDRV_BLOCK_DATA) {
1792 s->status = BLK_DATA;
1793 } else {
1794 s->status = s->target_has_backing ? BLK_BACKING_FILE : BLK_DATA;
1795 }
1796
1797 s->sector_next_status = sector_num + n;
1798 }
1799
1800 n = MIN(n, s->sector_next_status - sector_num);
1801 if (s->status == BLK_DATA) {
1802 n = MIN(n, s->buf_sectors);
1803 }
1804
1805 /* We need to write complete clusters for compressed images, so if an
1806 * unallocated area is shorter than that, we must consider the whole
1807 * cluster allocated. */
1808 if (s->compressed) {
1809 if (n < s->cluster_sectors) {
1810 n = MIN(s->cluster_sectors, s->total_sectors - sector_num);
1811 s->status = BLK_DATA;
1812 } else {
1813 n = QEMU_ALIGN_DOWN(n, s->cluster_sectors);
1814 }
1815 }
1816
1817 return n;
1818 }
1819
1820 static int coroutine_fn convert_co_read(ImgConvertState *s, int64_t sector_num,
1821 int nb_sectors, uint8_t *buf)
1822 {
1823 uint64_t single_read_until = 0;
1824 int n, ret;
1825
1826 assert(nb_sectors <= s->buf_sectors);
1827 while (nb_sectors > 0) {
1828 BlockBackend *blk;
1829 int src_cur;
1830 int64_t bs_sectors, src_cur_offset;
1831 uint64_t offset;
1832
1833 /* In the case of compression with multiple source files, we can get a
1834 * nb_sectors that spreads into the next part. So we must be able to
1835 * read across multiple BDSes for one convert_read() call. */
1836 convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1837 blk = s->src[src_cur];
1838 bs_sectors = s->src_sectors[src_cur];
1839
1840 offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
1841
1842 n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
1843 if (single_read_until > offset) {
1844 n = 1;
1845 }
1846
1847 ret = blk_co_pread(blk, offset, n << BDRV_SECTOR_BITS, buf, 0);
1848 if (ret < 0) {
1849 if (s->salvage) {
1850 if (n > 1) {
1851 single_read_until = offset + (n << BDRV_SECTOR_BITS);
1852 continue;
1853 } else {
1854 if (!s->quiet) {
1855 warn_report("error while reading offset %" PRIu64
1856 ": %s", offset, strerror(-ret));
1857 }
1858 memset(buf, 0, BDRV_SECTOR_SIZE);
1859 }
1860 } else {
1861 return ret;
1862 }
1863 }
1864
1865 sector_num += n;
1866 nb_sectors -= n;
1867 buf += n * BDRV_SECTOR_SIZE;
1868 }
1869
1870 return 0;
1871 }
1872
1873
1874 static int coroutine_fn convert_co_write(ImgConvertState *s, int64_t sector_num,
1875 int nb_sectors, uint8_t *buf,
1876 enum ImgConvertBlockStatus status)
1877 {
1878 int ret;
1879
1880 while (nb_sectors > 0) {
1881 int n = nb_sectors;
1882 BdrvRequestFlags flags = s->compressed ? BDRV_REQ_WRITE_COMPRESSED : 0;
1883
1884 switch (status) {
1885 case BLK_BACKING_FILE:
1886 /* If we have a backing file, leave clusters unallocated that are
1887 * unallocated in the source image, so that the backing file is
1888 * visible at the respective offset. */
1889 assert(s->target_has_backing);
1890 break;
1891
1892 case BLK_DATA:
1893 /* If we're told to keep the target fully allocated (-S 0) or there
1894 * is real non-zero data, we must write it. Otherwise we can treat
1895 * it as zero sectors.
1896 * Compressed clusters need to be written as a whole, so in that
1897 * case we can only save the write if the buffer is completely
1898 * zeroed. */
1899 if (!s->min_sparse ||
1900 (!s->compressed &&
1901 is_allocated_sectors_min(buf, n, &n, s->min_sparse,
1902 sector_num, s->alignment)) ||
1903 (s->compressed &&
1904 !buffer_is_zero(buf, n * BDRV_SECTOR_SIZE)))
1905 {
1906 ret = blk_co_pwrite(s->target, sector_num << BDRV_SECTOR_BITS,
1907 n << BDRV_SECTOR_BITS, buf, flags);
1908 if (ret < 0) {
1909 return ret;
1910 }
1911 break;
1912 }
1913 /* fall-through */
1914
1915 case BLK_ZERO:
1916 if (s->has_zero_init) {
1917 assert(!s->target_has_backing);
1918 break;
1919 }
1920 ret = blk_co_pwrite_zeroes(s->target,
1921 sector_num << BDRV_SECTOR_BITS,
1922 n << BDRV_SECTOR_BITS,
1923 BDRV_REQ_MAY_UNMAP);
1924 if (ret < 0) {
1925 return ret;
1926 }
1927 break;
1928 }
1929
1930 sector_num += n;
1931 nb_sectors -= n;
1932 buf += n * BDRV_SECTOR_SIZE;
1933 }
1934
1935 return 0;
1936 }
1937
1938 static int coroutine_fn convert_co_copy_range(ImgConvertState *s, int64_t sector_num,
1939 int nb_sectors)
1940 {
1941 int n, ret;
1942
1943 while (nb_sectors > 0) {
1944 BlockBackend *blk;
1945 int src_cur;
1946 int64_t bs_sectors, src_cur_offset;
1947 int64_t offset;
1948
1949 convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
1950 offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
1951 blk = s->src[src_cur];
1952 bs_sectors = s->src_sectors[src_cur];
1953
1954 n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
1955
1956 ret = blk_co_copy_range(blk, offset, s->target,
1957 sector_num << BDRV_SECTOR_BITS,
1958 n << BDRV_SECTOR_BITS, 0, 0);
1959 if (ret < 0) {
1960 return ret;
1961 }
1962
1963 sector_num += n;
1964 nb_sectors -= n;
1965 }
1966 return 0;
1967 }
1968
1969 static void coroutine_fn convert_co_do_copy(void *opaque)
1970 {
1971 ImgConvertState *s = opaque;
1972 uint8_t *buf = NULL;
1973 int ret, i;
1974 int index = -1;
1975
1976 for (i = 0; i < s->num_coroutines; i++) {
1977 if (s->co[i] == qemu_coroutine_self()) {
1978 index = i;
1979 break;
1980 }
1981 }
1982 assert(index >= 0);
1983
1984 s->running_coroutines++;
1985 buf = blk_blockalign(s->target, s->buf_sectors * BDRV_SECTOR_SIZE);
1986
1987 while (1) {
1988 int n;
1989 int64_t sector_num;
1990 enum ImgConvertBlockStatus status;
1991 bool copy_range;
1992
1993 qemu_co_mutex_lock(&s->lock);
1994 if (s->ret != -EINPROGRESS || s->sector_num >= s->total_sectors) {
1995 qemu_co_mutex_unlock(&s->lock);
1996 break;
1997 }
1998 WITH_GRAPH_RDLOCK_GUARD() {
1999 n = convert_iteration_sectors(s, s->sector_num);
2000 }
2001 if (n < 0) {
2002 qemu_co_mutex_unlock(&s->lock);
2003 s->ret = n;
2004 break;
2005 }
2006 /* save current sector and allocation status to local variables */
2007 sector_num = s->sector_num;
2008 status = s->status;
2009 if (!s->min_sparse && s->status == BLK_ZERO) {
2010 n = MIN(n, s->buf_sectors);
2011 }
2012 /* increment global sector counter so that other coroutines can
2013 * already continue reading beyond this request */
2014 s->sector_num += n;
2015 qemu_co_mutex_unlock(&s->lock);
2016
2017 if (status == BLK_DATA || (!s->min_sparse && status == BLK_ZERO)) {
2018 s->allocated_done += n;
2019 qemu_progress_print(100.0 * s->allocated_done /
2020 s->allocated_sectors, 0);
2021 }
2022
2023 retry:
2024 copy_range = s->copy_range && s->status == BLK_DATA;
2025 if (status == BLK_DATA && !copy_range) {
2026 ret = convert_co_read(s, sector_num, n, buf);
2027 if (ret < 0) {
2028 error_report("error while reading at byte %lld: %s",
2029 sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
2030 s->ret = ret;
2031 }
2032 } else if (!s->min_sparse && status == BLK_ZERO) {
2033 status = BLK_DATA;
2034 memset(buf, 0x00, n * BDRV_SECTOR_SIZE);
2035 }
2036
2037 if (s->wr_in_order) {
2038 /* keep writes in order */
2039 while (s->wr_offs != sector_num && s->ret == -EINPROGRESS) {
2040 s->wait_sector_num[index] = sector_num;
2041 qemu_coroutine_yield();
2042 }
2043 s->wait_sector_num[index] = -1;
2044 }
2045
2046 if (s->ret == -EINPROGRESS) {
2047 if (copy_range) {
2048 WITH_GRAPH_RDLOCK_GUARD() {
2049 ret = convert_co_copy_range(s, sector_num, n);
2050 }
2051 if (ret) {
2052 s->copy_range = false;
2053 goto retry;
2054 }
2055 } else {
2056 ret = convert_co_write(s, sector_num, n, buf, status);
2057 }
2058 if (ret < 0) {
2059 error_report("error while writing at byte %lld: %s",
2060 sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
2061 s->ret = ret;
2062 }
2063 }
2064
2065 if (s->wr_in_order) {
2066 /* reenter the coroutine that might have waited
2067 * for this write to complete */
2068 s->wr_offs = sector_num + n;
2069 for (i = 0; i < s->num_coroutines; i++) {
2070 if (s->co[i] && s->wait_sector_num[i] == s->wr_offs) {
2071 /*
2072 * A -> B -> A cannot occur because A has
2073 * s->wait_sector_num[i] == -1 during A -> B. Therefore
2074 * B will never enter A during this time window.
2075 */
2076 qemu_coroutine_enter(s->co[i]);
2077 break;
2078 }
2079 }
2080 }
2081 }
2082
2083 qemu_vfree(buf);
2084 s->co[index] = NULL;
2085 s->running_coroutines--;
2086 if (!s->running_coroutines && s->ret == -EINPROGRESS) {
2087 /* the convert job finished successfully */
2088 s->ret = 0;
2089 }
2090 }
2091
2092 static int convert_do_copy(ImgConvertState *s)
2093 {
2094 int ret, i, n;
2095 int64_t sector_num = 0;
2096
2097 /* Check whether we have zero initialisation or can get it efficiently */
2098 if (!s->has_zero_init && s->target_is_new && s->min_sparse &&
2099 !s->target_has_backing) {
2100 bdrv_graph_rdlock_main_loop();
2101 s->has_zero_init = bdrv_has_zero_init(blk_bs(s->target));
2102 bdrv_graph_rdunlock_main_loop();
2103 }
2104
2105 /* Allocate buffer for copied data. For compressed images, only one cluster
2106 * can be copied at a time. */
2107 if (s->compressed) {
2108 if (s->cluster_sectors <= 0 || s->cluster_sectors > s->buf_sectors) {
2109 error_report("invalid cluster size");
2110 return -EINVAL;
2111 }
2112 s->buf_sectors = s->cluster_sectors;
2113 }
2114
2115 while (sector_num < s->total_sectors) {
2116 bdrv_graph_rdlock_main_loop();
2117 n = convert_iteration_sectors(s, sector_num);
2118 bdrv_graph_rdunlock_main_loop();
2119 if (n < 0) {
2120 return n;
2121 }
2122 if (s->status == BLK_DATA || (!s->min_sparse && s->status == BLK_ZERO))
2123 {
2124 s->allocated_sectors += n;
2125 }
2126 sector_num += n;
2127 }
2128
2129 /* Do the copy */
2130 s->sector_next_status = 0;
2131 s->ret = -EINPROGRESS;
2132
2133 qemu_co_mutex_init(&s->lock);
2134 for (i = 0; i < s->num_coroutines; i++) {
2135 s->co[i] = qemu_coroutine_create(convert_co_do_copy, s);
2136 s->wait_sector_num[i] = -1;
2137 qemu_coroutine_enter(s->co[i]);
2138 }
2139
2140 while (s->running_coroutines) {
2141 main_loop_wait(false);
2142 }
2143
2144 if (s->compressed && !s->ret) {
2145 /* signal EOF to align */
2146 ret = blk_pwrite_compressed(s->target, 0, 0, NULL);
2147 if (ret < 0) {
2148 return ret;
2149 }
2150 }
2151
2152 return s->ret;
2153 }
2154
2155 /* Check that bitmaps can be copied, or output an error */
2156 static int convert_check_bitmaps(BlockDriverState *src, bool skip_broken)
2157 {
2158 BdrvDirtyBitmap *bm;
2159
2160 if (!bdrv_supports_persistent_dirty_bitmap(src)) {
2161 error_report("Source lacks bitmap support");
2162 return -1;
2163 }
2164 FOR_EACH_DIRTY_BITMAP(src, bm) {
2165 if (!bdrv_dirty_bitmap_get_persistence(bm)) {
2166 continue;
2167 }
2168 if (!skip_broken && bdrv_dirty_bitmap_inconsistent(bm)) {
2169 error_report("Cannot copy inconsistent bitmap '%s'",
2170 bdrv_dirty_bitmap_name(bm));
2171 error_printf("Try --skip-broken-bitmaps, or "
2172 "use 'qemu-img bitmap --remove' to delete it\n");
2173 return -1;
2174 }
2175 }
2176 return 0;
2177 }
2178
2179 static int convert_copy_bitmaps(BlockDriverState *src, BlockDriverState *dst,
2180 bool skip_broken)
2181 {
2182 BdrvDirtyBitmap *bm;
2183 Error *err = NULL;
2184
2185 FOR_EACH_DIRTY_BITMAP(src, bm) {
2186 const char *name;
2187
2188 if (!bdrv_dirty_bitmap_get_persistence(bm)) {
2189 continue;
2190 }
2191 name = bdrv_dirty_bitmap_name(bm);
2192 if (skip_broken && bdrv_dirty_bitmap_inconsistent(bm)) {
2193 warn_report("Skipping inconsistent bitmap '%s'", name);
2194 continue;
2195 }
2196 qmp_block_dirty_bitmap_add(dst->node_name, name,
2197 true, bdrv_dirty_bitmap_granularity(bm),
2198 true, true,
2199 true, !bdrv_dirty_bitmap_enabled(bm),
2200 &err);
2201 if (err) {
2202 error_reportf_err(err, "Failed to create bitmap %s: ", name);
2203 return -1;
2204 }
2205
2206 do_dirty_bitmap_merge(dst->node_name, name, src->node_name, name,
2207 &err);
2208 if (err) {
2209 error_reportf_err(err, "Failed to populate bitmap %s: ", name);
2210 qmp_block_dirty_bitmap_remove(dst->node_name, name, NULL);
2211 return -1;
2212 }
2213 }
2214
2215 return 0;
2216 }
2217
2218 #define MAX_BUF_SECTORS 32768
2219
2220 static void set_rate_limit(BlockBackend *blk, int64_t rate_limit)
2221 {
2222 ThrottleConfig cfg;
2223
2224 throttle_config_init(&cfg);
2225 cfg.buckets[THROTTLE_BPS_WRITE].avg = rate_limit;
2226
2227 blk_io_limits_enable(blk, CONVERT_THROTTLE_GROUP);
2228 blk_set_io_limits(blk, &cfg);
2229 }
2230
2231 static int img_convert(const img_cmd_t *ccmd, int argc, char **argv)
2232 {
2233 int c, bs_i, flags, src_flags = BDRV_O_NO_SHARE;
2234 const char *fmt = NULL, *out_fmt = NULL, *cache = "unsafe",
2235 *src_cache = BDRV_DEFAULT_CACHE, *out_baseimg = NULL,
2236 *out_filename, *out_baseimg_param, *snapshot_name = NULL,
2237 *backing_fmt = NULL;
2238 BlockDriver *drv = NULL, *proto_drv = NULL;
2239 BlockDriverInfo bdi;
2240 BlockDriverState *out_bs;
2241 QemuOpts *opts = NULL, *sn_opts = NULL;
2242 QemuOptsList *create_opts = NULL;
2243 QDict *open_opts = NULL;
2244 char *options = NULL;
2245 Error *local_err = NULL;
2246 bool writethrough, src_writethrough, image_opts = false,
2247 skip_create = false, progress = false, tgt_image_opts = false;
2248 int64_t ret = -EINVAL;
2249 bool force_share = false;
2250 bool explict_min_sparse = false;
2251 bool bitmaps = false;
2252 bool skip_broken = false;
2253 int64_t rate_limit = 0;
2254
2255 ImgConvertState s = (ImgConvertState) {
2256 /* Need at least 4k of zeros for sparse detection */
2257 .min_sparse = 8,
2258 .copy_range = false,
2259 .buf_sectors = IO_BUF_SIZE / BDRV_SECTOR_SIZE,
2260 .wr_in_order = true,
2261 .num_coroutines = 8,
2262 };
2263
2264 for(;;) {
2265 static const struct option long_options[] = {
2266 {"help", no_argument, 0, 'h'},
2267 {"source-format", required_argument, 0, 'f'},
2268 /*
2269 * XXX: historic --image-opts acts on source file only,
2270 * it seems better to have it affect both source and target,
2271 * and have separate --source-image-opts for source,
2272 * but this might break existing setups.
2273 */
2274 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
2275 {"source-cache", required_argument, 0, 'T'},
2276 {"snapshot", required_argument, 0, 'l'},
2277 {"bitmaps", no_argument, 0, OPTION_BITMAPS},
2278 {"skip-broken-bitmaps", no_argument, 0, OPTION_SKIP_BROKEN},
2279 {"salvage", no_argument, 0, OPTION_SALVAGE},
2280 {"target-format", required_argument, 0, 'O'},
2281 {"target-image-opts", no_argument, 0, OPTION_TARGET_IMAGE_OPTS},
2282 {"target-format-options", required_argument, 0, 'o'},
2283 {"target-cache", required_argument, 0, 't'},
2284 {"backing", required_argument, 0, 'b'},
2285 {"backing-format", required_argument, 0, 'F'},
2286 {"sparse-size", required_argument, 0, 'S'},
2287 {"no-create", no_argument, 0, 'n'},
2288 {"target-is-zero", no_argument, 0, OPTION_TARGET_IS_ZERO},
2289 {"force-share", no_argument, 0, 'U'},
2290 {"rate-limit", required_argument, 0, 'r'},
2291 {"parallel", required_argument, 0, 'm'},
2292 {"oob-writes", no_argument, 0, 'W'},
2293 {"copy-range-offloading", no_argument, 0, 'C'},
2294 {"progress", no_argument, 0, 'p'},
2295 {"quiet", no_argument, 0, 'q'},
2296 {"object", required_argument, 0, OPTION_OBJECT},
2297 {0, 0, 0, 0}
2298 };
2299 c = getopt_long(argc, argv, "hf:O:b:B:CcF:o:l:S:pt:T:nm:WUr:q",
2300 long_options, NULL);
2301 if (c == -1) {
2302 break;
2303 }
2304 switch (c) {
2305 case 'h':
2306 cmd_help(ccmd, "[-f SRC_FMT | --image-opts] [-T SRC_CACHE]\n"
2307 " [-l SNAPSHOT] [--bitmaps [--skip-broken-bitmaps]] [--salvage]\n"
2308 " [-O TGT_FMT | --target-image-opts] [-o TGT_FMT_OPTS] [-t TGT_CACHE]\n"
2309 " [-b BACKING_FILE [-F BACKING_FMT]] [-S SPARSE_SIZE]\n"
2310 " [-n] [--target-is-zero] [-c]\n"
2311 " [-U] [-r RATE] [-m NUM_PARALLEL] [-W] [-C] [-p] [-q] [--object OBJDEF]\n"
2312 " SRC_FILE [SRC_FILE2...] TGT_FILE\n"
2313 ,
2314 " -f, --source-format SRC_FMT\n"
2315 " specify format of all SRC_FILEs explicitly (default: probing is used)\n"
2316 " --image-opts\n"
2317 " treat each SRC_FILE as an option string (key=value,...), not a file name\n"
2318 " (incompatible with -f|--source-format)\n"
2319 " -T, --source-cache SRC_CACHE\n"
2320 " source image(s) cache mode (" BDRV_DEFAULT_CACHE ")\n"
2321 " -l, --snapshot SNAPSHOT\n"
2322 " specify source snapshot\n"
2323 " --bitmaps\n"
2324 " also copy any persistent bitmaps present in source\n"
2325 " --skip-broken-bitmaps\n"
2326 " skip (do not error out) any broken bitmaps\n"
2327 " --salvage\n"
2328 " ignore errors on input (convert unreadable areas to zeros)\n"
2329 " -O, --target-format TGT_FMT\n"
2330 " specify TGT_FILE image format (default: raw)\n"
2331 " --target-image-opts\n"
2332 " treat TGT_FILE as an option string (key=value,...), not a file name\n"
2333 " (incompatible with -O|--target-format)\n"
2334 " -o, --target-format-options TGT_FMT_OPTS\n"
2335 " TGT_FMT-specific options\n"
2336 " -t, --target-cache TGT_CACHE\n"
2337 " cache mode when opening output image (default: unsafe)\n"
2338 " -b, --backing BACKING_FILE (was -B in <= 10.0)\n"
2339 " create target image to be a CoW on top of BACKING_FILE\n"
2340 " -F, --backing-format BACKING_FMT\n" /* -B used for -b in <=10.0 */
2341 " specify BACKING_FILE image format explicitly (default: probing is used)\n"
2342 " -S, --sparse-size SPARSE_SIZE[bkKMGTPE]\n"
2343 " specify number of consecutive zero bytes to treat as a gap on output\n"
2344 " (rounded down to nearest 512 bytes), with optional multiplier suffix\n"
2345 " -n, --no-create\n"
2346 " omit target volume creation (e.g. on rbd)\n"
2347 " --target-is-zero\n"
2348 " indicates that the target volume is pre-zeroed\n"
2349 " -c, --compress\n"
2350 " create compressed output image (qcow and qcow2 formats only)\n"
2351 " -U, --force-share\n"
2352 " open images in shared mode for concurrent access\n"
2353 " -r, --rate-limit RATE\n"
2354 " I/O rate limit, in bytes per second\n"
2355 " -m, --parallel NUM_PARALLEL\n"
2356 " specify parallelism (default: 8)\n"
2357 " -C, --copy-range-offloading\n"
2358 " try to use copy offloading\n"
2359 " -W, --oob-writes\n"
2360 " enable out-of-order writes to improve performance\n"
2361 " -p, --progress\n"
2362 " display progress information\n"
2363 " -q, --quiet\n"
2364 " quiet mode (produce only error messages if any)\n"
2365 " --object OBJDEF\n"
2366 " defines QEMU user-creatable object\n"
2367 " SRC_FILE...\n"
2368 " one or more source image file names,\n"
2369 " or option strings (key=value,..) with --source-image-opts\n"
2370 " TGT_FILE\n"
2371 " target (output) image file name,\n"
2372 " or option string (key=value,..) with --target-image-opts\n"
2373 );
2374 break;
2375 case 'f':
2376 fmt = optarg;
2377 break;
2378 case OPTION_IMAGE_OPTS:
2379 image_opts = true;
2380 break;
2381 case 'T':
2382 src_cache = optarg;
2383 break;
2384 case 'l':
2385 if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
2386 sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
2387 optarg, false);
2388 if (!sn_opts) {
2389 error_report("Failed in parsing snapshot param '%s'",
2390 optarg);
2391 goto fail_getopt;
2392 }
2393 } else {
2394 snapshot_name = optarg;
2395 }
2396 break;
2397 case OPTION_BITMAPS:
2398 bitmaps = true;
2399 break;
2400 case OPTION_SKIP_BROKEN:
2401 skip_broken = true;
2402 break;
2403 case OPTION_SALVAGE:
2404 s.salvage = true;
2405 break;
2406 case 'O':
2407 out_fmt = optarg;
2408 break;
2409 case OPTION_TARGET_IMAGE_OPTS:
2410 tgt_image_opts = true;
2411 break;
2412 case 'o':
2413 if (accumulate_options(&options, optarg) < 0) {
2414 goto fail_getopt;
2415 }
2416 break;
2417 case 't':
2418 cache = optarg;
2419 break;
2420 case 'B': /* <=10.0 */
2421 case 'b':
2422 out_baseimg = optarg;
2423 break;
2424 case 'F': /* can't use -B as it used as -b in <=10.0 */
2425 backing_fmt = optarg;
2426 break;
2427 case 'S':
2428 {
2429 int64_t sval;
2430
2431 sval = cvtnum("buffer size for sparse output", optarg, true);
2432 if (sval < 0) {
2433 goto fail_getopt;
2434 } else if (!QEMU_IS_ALIGNED(sval, BDRV_SECTOR_SIZE) ||
2435 sval / BDRV_SECTOR_SIZE > MAX_BUF_SECTORS) {
2436 error_report("Invalid buffer size for sparse output specified. "
2437 "Valid sizes are multiples of %llu up to %llu. Select "
2438 "0 to disable sparse detection (fully allocates output).",
2439 BDRV_SECTOR_SIZE, MAX_BUF_SECTORS * BDRV_SECTOR_SIZE);
2440 goto fail_getopt;
2441 }
2442
2443 s.min_sparse = sval / BDRV_SECTOR_SIZE;
2444 explict_min_sparse = true;
2445 break;
2446 }
2447 case 'n':
2448 skip_create = true;
2449 break;
2450 case OPTION_TARGET_IS_ZERO:
2451 /*
2452 * The user asserting that the target is blank has the
2453 * same effect as the target driver supporting zero
2454 * initialisation.
2455 */
2456 s.has_zero_init = true;
2457 break;
2458 case 'c':
2459 s.compressed = true;
2460 break;
2461 case 'U':
2462 force_share = true;
2463 break;
2464 case 'r':
2465 rate_limit = cvtnum("rate limit", optarg, true);
2466 if (rate_limit < 0) {
2467 goto fail_getopt;
2468 }
2469 break;
2470 case 'm':
2471 s.num_coroutines = cvtnum_full("number of coroutines", optarg,
2472 false, 1, MAX_COROUTINES);
2473 if (s.num_coroutines < 0) {
2474 goto fail_getopt;
2475 }
2476 break;
2477 case 'W':
2478 s.wr_in_order = false;
2479 break;
2480 case 'C':
2481 s.copy_range = true;
2482 break;
2483 case 'p':
2484 progress = true;
2485 break;
2486 case 'q':
2487 s.quiet = true;
2488 break;
2489 case OPTION_OBJECT:
2490 user_creatable_process_cmdline(optarg);
2491 break;
2492 default:
2493 tryhelp(argv[0]);
2494 }
2495 }
2496
2497 if (!out_fmt && !tgt_image_opts) {
2498 out_fmt = "raw";
2499 }
2500
2501 if (skip_broken && !bitmaps) {
2502 error_report("Use of --skip-broken-bitmaps requires --bitmaps");
2503 goto fail_getopt;
2504 }
2505
2506 if (s.compressed && s.copy_range) {
2507 error_report("Cannot enable copy offloading when -c is used");
2508 goto fail_getopt;
2509 }
2510
2511 if (explict_min_sparse && s.copy_range) {
2512 error_report("Cannot enable copy offloading when -S is used");
2513 goto fail_getopt;
2514 }
2515
2516 if (s.copy_range && s.salvage) {
2517 error_report("Cannot use copy offloading in salvaging mode");
2518 goto fail_getopt;
2519 }
2520
2521 if (tgt_image_opts && !skip_create) {
2522 error_report("--target-image-opts requires use of -n flag");
2523 goto fail_getopt;
2524 }
2525
2526 if (skip_create && options) {
2527 error_report("-o has no effect when skipping image creation");
2528 goto fail_getopt;
2529 }
2530
2531 if (s.has_zero_init && !skip_create) {
2532 error_report("--target-is-zero requires use of -n flag");
2533 goto fail_getopt;
2534 }
2535
2536 s.src_num = argc - optind - 1;
2537 out_filename = s.src_num >= 1 ? argv[argc - 1] : NULL;
2538
2539 if (options && has_help_option(options)) {
2540 if (out_fmt) {
2541 ret = print_block_option_help(out_filename, out_fmt);
2542 goto fail_getopt;
2543 } else {
2544 error_report("Option help requires a format be specified");
2545 goto fail_getopt;
2546 }
2547 }
2548
2549 if (s.src_num < 1) {
2550 error_report("Must specify image file name");
2551 goto fail_getopt;
2552 }
2553
2554 /* ret is still -EINVAL until here */
2555 ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
2556 if (ret < 0) {
2557 error_report("Invalid source cache option: %s", src_cache);
2558 goto fail_getopt;
2559 }
2560
2561 /* Initialize before goto out */
2562 if (s.quiet) {
2563 progress = false;
2564 }
2565 qemu_progress_init(progress, 1.0);
2566 qemu_progress_print(0, 100);
2567
2568 s.src = g_new0(BlockBackend *, s.src_num);
2569 s.src_sectors = g_new(int64_t, s.src_num);
2570 s.src_alignment = g_new(int, s.src_num);
2571
2572 for (bs_i = 0; bs_i < s.src_num; bs_i++) {
2573 BlockDriverState *src_bs;
2574 s.src[bs_i] = img_open(image_opts, argv[optind + bs_i],
2575 fmt, src_flags, src_writethrough, s.quiet,
2576 force_share);
2577 if (!s.src[bs_i]) {
2578 ret = -1;
2579 goto out;
2580 }
2581 s.src_sectors[bs_i] = blk_nb_sectors(s.src[bs_i]);
2582 if (s.src_sectors[bs_i] < 0) {
2583 error_report("Could not get size of %s: %s",
2584 argv[optind + bs_i], strerror(-s.src_sectors[bs_i]));
2585 ret = -1;
2586 goto out;
2587 }
2588 src_bs = blk_bs(s.src[bs_i]);
2589 s.src_alignment[bs_i] = DIV_ROUND_UP(src_bs->bl.request_alignment,
2590 BDRV_SECTOR_SIZE);
2591 if (!bdrv_get_info(src_bs, &bdi)) {
2592 s.src_alignment[bs_i] = MAX(s.src_alignment[bs_i],
2593 bdi.cluster_size / BDRV_SECTOR_SIZE);
2594 }
2595 s.total_sectors += s.src_sectors[bs_i];
2596 }
2597
2598 if (sn_opts) {
2599 bdrv_snapshot_load_tmp(blk_bs(s.src[0]),
2600 qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
2601 qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
2602 &local_err);
2603 } else if (snapshot_name != NULL) {
2604 if (s.src_num > 1) {
2605 error_report("No support for concatenating multiple snapshot");
2606 ret = -1;
2607 goto out;
2608 }
2609
2610 bdrv_snapshot_load_tmp_by_id_or_name(blk_bs(s.src[0]), snapshot_name,
2611 &local_err);
2612 }
2613 if (local_err) {
2614 error_reportf_err(local_err, "Failed to load snapshot: ");
2615 ret = -1;
2616 goto out;
2617 }
2618
2619 if (!skip_create) {
2620 /* Find driver and parse its options */
2621 drv = bdrv_find_format(out_fmt);
2622 if (!drv) {
2623 error_report("Unknown file format '%s'", out_fmt);
2624 ret = -1;
2625 goto out;
2626 }
2627
2628 proto_drv = bdrv_find_protocol(out_filename, true, &local_err);
2629 if (!proto_drv) {
2630 error_report_err(local_err);
2631 ret = -1;
2632 goto out;
2633 }
2634
2635 if (!drv->create_opts) {
2636 error_report("Format driver '%s' does not support image creation",
2637 drv->format_name);
2638 ret = -1;
2639 goto out;
2640 }
2641
2642 if (!proto_drv->create_opts) {
2643 error_report("Protocol driver '%s' does not support image creation",
2644 proto_drv->format_name);
2645 ret = -1;
2646 goto out;
2647 }
2648
2649 create_opts = qemu_opts_append(create_opts, drv->create_opts);
2650 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
2651
2652 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
2653 if (options) {
2654 if (!qemu_opts_do_parse(opts, options, NULL, &local_err)) {
2655 error_report_err(local_err);
2656 ret = -1;
2657 goto out;
2658 }
2659 }
2660
2661 qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
2662 s.total_sectors * BDRV_SECTOR_SIZE, &error_abort);
2663 ret = add_old_style_options(out_fmt, opts, out_baseimg, backing_fmt);
2664 if (ret < 0) {
2665 goto out;
2666 }
2667 }
2668
2669 /* Get backing file name if -o backing_file was used */
2670 out_baseimg_param = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
2671 if (out_baseimg_param) {
2672 out_baseimg = out_baseimg_param;
2673 }
2674 s.target_has_backing = (bool) out_baseimg;
2675
2676 if (s.has_zero_init && s.target_has_backing) {
2677 error_report("Cannot use --target-is-zero when the destination "
2678 "image has a backing file");
2679 goto out;
2680 }
2681
2682 if (s.src_num > 1 && out_baseimg) {
2683 error_report("Having a backing file for the target makes no sense when "
2684 "concatenating multiple input images");
2685 ret = -1;
2686 goto out;
2687 }
2688
2689 if (out_baseimg_param) {
2690 if (!qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT)) {
2691 error_report("Use of backing file requires explicit "
2692 "backing format");
2693 ret = -1;
2694 goto out;
2695 }
2696 }
2697
2698 /* Check if compression is supported */
2699 if (s.compressed) {
2700 bool encryption =
2701 qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, false);
2702 const char *encryptfmt =
2703 qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT);
2704 const char *preallocation =
2705 qemu_opt_get(opts, BLOCK_OPT_PREALLOC);
2706
2707 if (drv && !block_driver_can_compress(drv)) {
2708 error_report("Compression not supported for this file format");
2709 ret = -1;
2710 goto out;
2711 }
2712
2713 if (encryption || encryptfmt) {
2714 error_report("Compression and encryption not supported at "
2715 "the same time");
2716 ret = -1;
2717 goto out;
2718 }
2719
2720 if (preallocation
2721 && strcmp(preallocation, "off"))
2722 {
2723 error_report("Compression and preallocation not supported at "
2724 "the same time");
2725 ret = -1;
2726 goto out;
2727 }
2728 }
2729
2730 /* Determine if bitmaps need copying */
2731 if (bitmaps) {
2732 if (s.src_num > 1) {
2733 error_report("Copying bitmaps only possible with single source");
2734 ret = -1;
2735 goto out;
2736 }
2737 ret = convert_check_bitmaps(blk_bs(s.src[0]), skip_broken);
2738 if (ret < 0) {
2739 goto out;
2740 }
2741 }
2742
2743 /*
2744 * The later open call will need any decryption secrets, and
2745 * bdrv_create() will purge "opts", so extract them now before
2746 * they are lost.
2747 */
2748 if (!skip_create) {
2749 open_opts = qdict_new();
2750 qemu_opt_foreach(opts, img_add_key_secrets, open_opts, &error_abort);
2751
2752 /* Create the new image */
2753 ret = bdrv_create(drv, out_filename, opts, &local_err);
2754 if (ret < 0) {
2755 error_reportf_err(local_err, "%s: error while converting %s: ",
2756 out_filename, out_fmt);
2757 goto out;
2758 }
2759 }
2760
2761 s.target_is_new = !skip_create;
2762
2763 flags = s.min_sparse ? (BDRV_O_RDWR | BDRV_O_UNMAP) : BDRV_O_RDWR;
2764 ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
2765 if (ret < 0) {
2766 error_report("Invalid cache option: %s", cache);
2767 goto out;
2768 }
2769
2770 if (flags & BDRV_O_NOCACHE) {
2771 /*
2772 * If we open the target with O_DIRECT, it may be necessary to
2773 * extend its size to align to the physical sector size.
2774 */
2775 flags |= BDRV_O_RESIZE;
2776 }
2777
2778 if (skip_create) {
2779 s.target = img_open(tgt_image_opts, out_filename, out_fmt,
2780 flags, writethrough, s.quiet, false);
2781 } else {
2782 /* TODO ultimately we should allow --target-image-opts
2783 * to be used even when -n is not given.
2784 * That has to wait for bdrv_create to be improved
2785 * to allow filenames in option syntax
2786 */
2787 s.target = img_open_file(out_filename, open_opts, out_fmt,
2788 flags, writethrough, s.quiet, false);
2789 open_opts = NULL; /* blk_new_open will have freed it */
2790 }
2791 if (!s.target) {
2792 ret = -1;
2793 goto out;
2794 }
2795 out_bs = blk_bs(s.target);
2796
2797 if (bitmaps && !bdrv_supports_persistent_dirty_bitmap(out_bs)) {
2798 error_report("Format driver '%s' does not support bitmaps",
2799 out_bs->drv->format_name);
2800 ret = -1;
2801 goto out;
2802 }
2803
2804 if (s.compressed && !block_driver_can_compress(out_bs->drv)) {
2805 error_report("Compression not supported for this file format");
2806 ret = -1;
2807 goto out;
2808 }
2809
2810 /* increase bufsectors from the default 4096 (2M) if opt_transfer
2811 * or discard_alignment of the out_bs is greater. Limit to
2812 * MAX_BUF_SECTORS as maximum which is currently 32768 (16MB). */
2813 s.buf_sectors = MIN(MAX_BUF_SECTORS,
2814 MAX(s.buf_sectors,
2815 MAX(out_bs->bl.opt_transfer >> BDRV_SECTOR_BITS,
2816 out_bs->bl.pdiscard_alignment >>
2817 BDRV_SECTOR_BITS)));
2818
2819 /* try to align the write requests to the destination to avoid unnecessary
2820 * RMW cycles. */
2821 s.alignment = MAX(pow2floor(s.min_sparse),
2822 DIV_ROUND_UP(out_bs->bl.request_alignment,
2823 BDRV_SECTOR_SIZE));
2824 assert(is_power_of_2(s.alignment));
2825
2826 if (skip_create) {
2827 int64_t output_sectors = blk_nb_sectors(s.target);
2828 if (output_sectors < 0) {
2829 error_report("unable to get output image length: %s",
2830 strerror(-output_sectors));
2831 ret = -1;
2832 goto out;
2833 } else if (output_sectors < s.total_sectors) {
2834 error_report("output file is smaller than input file");
2835 ret = -1;
2836 goto out;
2837 }
2838 }
2839
2840 if (s.target_has_backing && s.target_is_new) {
2841 /* Errors are treated as "backing length unknown" (which means
2842 * s.target_backing_sectors has to be negative, which it will
2843 * be automatically). The backing file length is used only
2844 * for optimizations, so such a case is not fatal. */
2845 bdrv_graph_rdlock_main_loop();
2846 s.target_backing_sectors =
2847 bdrv_nb_sectors(bdrv_backing_chain_next(out_bs));
2848 bdrv_graph_rdunlock_main_loop();
2849 } else {
2850 s.target_backing_sectors = -1;
2851 }
2852
2853 ret = bdrv_get_info(out_bs, &bdi);
2854 if (ret < 0) {
2855 if (s.compressed) {
2856 error_report("could not get block driver info");
2857 goto out;
2858 }
2859 } else {
2860 s.compressed = s.compressed || bdi.needs_compressed_writes;
2861 s.cluster_sectors = bdi.cluster_size / BDRV_SECTOR_SIZE;
2862 }
2863
2864 if (rate_limit) {
2865 set_rate_limit(s.target, rate_limit);
2866 }
2867
2868 ret = convert_do_copy(&s);
2869
2870 /* Now copy the bitmaps */
2871 if (bitmaps && ret == 0) {
2872 ret = convert_copy_bitmaps(blk_bs(s.src[0]), out_bs, skip_broken);
2873 }
2874
2875 out:
2876 if (!ret) {
2877 qemu_progress_print(100, 0);
2878 }
2879 qemu_progress_end();
2880 qemu_opts_del(opts);
2881 qemu_opts_free(create_opts);
2882 qobject_unref(open_opts);
2883 blk_unref(s.target);
2884 if (s.src) {
2885 for (bs_i = 0; bs_i < s.src_num; bs_i++) {
2886 blk_unref(s.src[bs_i]);
2887 }
2888 g_free(s.src);
2889 }
2890 g_free(s.src_sectors);
2891 g_free(s.src_alignment);
2892 fail_getopt:
2893 qemu_opts_del(sn_opts);
2894 g_free(options);
2895
2896 return !!ret;
2897 }
2898
2899
2900 static void dump_snapshots(BlockDriverState *bs)
2901 {
2902 QEMUSnapshotInfo *sn_tab, *sn;
2903 int nb_sns, i;
2904
2905 nb_sns = bdrv_snapshot_list(bs, &sn_tab);
2906 if (nb_sns <= 0)
2907 return;
2908 printf("Snapshot list:\n");
2909 bdrv_snapshot_dump(NULL);
2910 printf("\n");
2911 for(i = 0; i < nb_sns; i++) {
2912 sn = &sn_tab[i];
2913 bdrv_snapshot_dump(sn);
2914 printf("\n");
2915 }
2916 g_free(sn_tab);
2917 }
2918
2919 static void dump_json_block_graph_info_list(BlockGraphInfoList *list)
2920 {
2921 GString *str;
2922 QObject *obj;
2923 Visitor *v = qobject_output_visitor_new(&obj);
2924
2925 visit_type_BlockGraphInfoList(v, NULL, &list, &error_abort);
2926 visit_complete(v, &obj);
2927 str = qobject_to_json_pretty(obj, true);
2928 assert(str != NULL);
2929 printf("%s\n", str->str);
2930 qobject_unref(obj);
2931 visit_free(v);
2932 g_string_free(str, true);
2933 }
2934
2935 static void dump_json_block_graph_info(BlockGraphInfo *info)
2936 {
2937 GString *str;
2938 QObject *obj;
2939 Visitor *v = qobject_output_visitor_new(&obj);
2940
2941 visit_type_BlockGraphInfo(v, NULL, &info, &error_abort);
2942 visit_complete(v, &obj);
2943 str = qobject_to_json_pretty(obj, true);
2944 assert(str != NULL);
2945 printf("%s\n", str->str);
2946 qobject_unref(obj);
2947 visit_free(v);
2948 g_string_free(str, true);
2949 }
2950
2951 static void dump_human_image_info(BlockGraphInfo *info, int indentation,
2952 const char *path)
2953 {
2954 BlockChildInfoList *children_list;
2955
2956 bdrv_node_info_dump(qapi_BlockGraphInfo_base(info), indentation,
2957 info->children == NULL);
2958
2959 for (children_list = info->children; children_list;
2960 children_list = children_list->next)
2961 {
2962 BlockChildInfo *child = children_list->value;
2963 g_autofree char *child_path = NULL;
2964
2965 printf("%*sChild node '%s%s':\n",
2966 indentation * 4, "", path, child->name);
2967 child_path = g_strdup_printf("%s%s/", path, child->name);
2968 dump_human_image_info(child->info, indentation + 1, child_path);
2969 }
2970 }
2971
2972 static void dump_human_image_info_list(BlockGraphInfoList *list)
2973 {
2974 BlockGraphInfoList *elem;
2975 bool delim = false;
2976
2977 for (elem = list; elem; elem = elem->next) {
2978 if (delim) {
2979 printf("\n");
2980 }
2981 delim = true;
2982
2983 dump_human_image_info(elem->value, 0, "/");
2984 }
2985 }
2986
2987 static gboolean str_equal_func(gconstpointer a, gconstpointer b)
2988 {
2989 return strcmp(a, b) == 0;
2990 }
2991
2992 /**
2993 * Open an image file chain and return an BlockGraphInfoList
2994 *
2995 * @filename: topmost image filename
2996 * @fmt: topmost image format (may be NULL to autodetect)
2997 * @chain: true - enumerate entire backing file chain
2998 * false - only topmost image file
2999 *
3000 * Returns a list of BlockNodeInfo objects or NULL if there was an error
3001 * opening an image file. If there was an error a message will have been
3002 * printed to stderr.
3003 */
3004 static BlockGraphInfoList *collect_image_info_list(bool image_opts,
3005 const char *filename,
3006 const char *fmt,
3007 const char *cache,
3008 bool chain, bool limits,
3009 bool force_share)
3010 {
3011 BlockGraphInfoList *head = NULL;
3012 BlockGraphInfoList **tail = &head;
3013 GHashTable *filenames;
3014 Error *err = NULL;
3015 int cache_flags = 0;
3016 bool writethrough = false;
3017 int ret;
3018
3019 ret = bdrv_parse_cache_mode(cache, &cache_flags, &writethrough);
3020 if (ret < 0) {
3021 error_report("Invalid cache option: %s", cache);
3022 return NULL;
3023 }
3024
3025 filenames = g_hash_table_new_full(g_str_hash, str_equal_func, NULL, NULL);
3026
3027 while (filename) {
3028 BlockBackend *blk;
3029 BlockDriverState *bs;
3030 BlockGraphInfo *info;
3031
3032 if (g_hash_table_lookup_extended(filenames, filename, NULL, NULL)) {
3033 error_report("Backing file '%s' creates an infinite loop.",
3034 filename);
3035 goto err;
3036 }
3037 g_hash_table_insert(filenames, (gpointer)filename, NULL);
3038
3039 blk = img_open(image_opts, filename, fmt,
3040 BDRV_O_NO_BACKING | BDRV_O_NO_IO | cache_flags,
3041 writethrough, false, force_share);
3042 if (!blk) {
3043 goto err;
3044 }
3045 bs = blk_bs(blk);
3046
3047 /*
3048 * Note that the returned BlockGraphInfo object will not have
3049 * information about this image's backing node, because we have opened
3050 * it with BDRV_O_NO_BACKING. Printing this object will therefore not
3051 * duplicate the backing chain information that we obtain by walking
3052 * the chain manually here.
3053 */
3054 bdrv_graph_rdlock_main_loop();
3055 bdrv_query_block_graph_info(bs, &info, limits, &err);
3056 bdrv_graph_rdunlock_main_loop();
3057
3058 if (err) {
3059 error_report_err(err);
3060 blk_unref(blk);
3061 goto err;
3062 }
3063
3064 QAPI_LIST_APPEND(tail, info);
3065
3066 blk_unref(blk);
3067
3068 /* Clear parameters that only apply to the topmost image */
3069 filename = fmt = NULL;
3070 image_opts = false;
3071
3072 if (chain) {
3073 if (info->full_backing_filename) {
3074 filename = info->full_backing_filename;
3075 } else if (info->backing_filename) {
3076 error_report("Could not determine absolute backing filename,"
3077 " but backing filename '%s' present",
3078 info->backing_filename);
3079 goto err;
3080 }
3081 if (info->backing_filename_format) {
3082 fmt = info->backing_filename_format;
3083 }
3084 }
3085 }
3086 g_hash_table_destroy(filenames);
3087 return head;
3088
3089 err:
3090 qapi_free_BlockGraphInfoList(head);
3091 g_hash_table_destroy(filenames);
3092 return NULL;
3093 }
3094
3095 static int img_info(const img_cmd_t *ccmd, int argc, char **argv)
3096 {
3097 int c;
3098 OutputFormat output_format = OFORMAT_HUMAN;
3099 bool chain = false;
3100 const char *filename, *fmt;
3101 const char *cache = BDRV_DEFAULT_CACHE;
3102 BlockGraphInfoList *list;
3103 bool image_opts = false;
3104 bool force_share = false;
3105 bool limits = false;
3106
3107 fmt = NULL;
3108 for(;;) {
3109 static const struct option long_options[] = {
3110 {"help", no_argument, 0, 'h'},
3111 {"format", required_argument, 0, 'f'},
3112 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3113 {"backing-chain", no_argument, 0, OPTION_BACKING_CHAIN},
3114 {"cache", required_argument, 0, 't'},
3115 {"force-share", no_argument, 0, 'U'},
3116 {"limits", no_argument, 0, OPTION_LIMITS},
3117 {"output", required_argument, 0, OPTION_OUTPUT},
3118 {"object", required_argument, 0, OPTION_OBJECT},
3119 {0, 0, 0, 0}
3120 };
3121 c = getopt_long(argc, argv, "hf:t:U", long_options, NULL);
3122 if (c == -1) {
3123 break;
3124 }
3125 switch(c) {
3126 case 'h':
3127 cmd_help(ccmd, "[-f FMT | --image-opts] [--backing-chain] [-U]\n"
3128 " [--output human|json] [--object OBJDEF] FILE\n"
3129 ,
3130 " -f, --format FMT\n"
3131 " specify FILE image format explicitly (default: probing is used)\n"
3132 " --image-opts\n"
3133 " treat FILE as an option string (key=value,..), not a file name\n"
3134 " (incompatible with -f|--format)\n"
3135 " --backing-chain\n"
3136 " display information about the backing chain for copy-on-write overlays\n"
3137 " -t, --cache CACHE\n"
3138 " cache mode for FILE (default: " BDRV_DEFAULT_CACHE ")\n"
3139 " -U, --force-share\n"
3140 " open image in shared mode for concurrent access\n"
3141 " --limits\n"
3142 " show detected block limits (may depend on options, e.g. cache mode)\n"
3143 " --output human|json\n"
3144 " specify output format (default: human)\n"
3145 " --object OBJDEF\n"
3146 " defines QEMU user-creatable object\n"
3147 " FILE\n"
3148 " name of the image file, or option string (key=value,..)\n"
3149 " with --image-opts, to operate on\n"
3150 );
3151 break;
3152 case 'f':
3153 fmt = optarg;
3154 break;
3155 case OPTION_IMAGE_OPTS:
3156 image_opts = true;
3157 break;
3158 case OPTION_BACKING_CHAIN:
3159 chain = true;
3160 break;
3161 case 't':
3162 cache = optarg;
3163 break;
3164 case 'U':
3165 force_share = true;
3166 break;
3167 case OPTION_LIMITS:
3168 limits = true;
3169 break;
3170 case OPTION_OUTPUT:
3171 output_format = parse_output_format(argv[0], optarg);
3172 break;
3173 case OPTION_OBJECT:
3174 user_creatable_process_cmdline(optarg);
3175 break;
3176 default:
3177 tryhelp(argv[0]);
3178 }
3179 }
3180 if (optind != argc - 1) {
3181 error_exit(argv[0], "Expecting one image file name");
3182 }
3183 filename = argv[optind++];
3184
3185 list = collect_image_info_list(image_opts, filename, fmt, cache, chain,
3186 limits, force_share);
3187 if (!list) {
3188 return 1;
3189 }
3190
3191 switch (output_format) {
3192 case OFORMAT_HUMAN:
3193 dump_human_image_info_list(list);
3194 break;
3195 case OFORMAT_JSON:
3196 if (chain) {
3197 dump_json_block_graph_info_list(list);
3198 } else {
3199 dump_json_block_graph_info(list->value);
3200 }
3201 break;
3202 }
3203
3204 qapi_free_BlockGraphInfoList(list);
3205 return 0;
3206 }
3207
3208 static int dump_map_entry(OutputFormat output_format, MapEntry *e,
3209 MapEntry *next)
3210 {
3211 switch (output_format) {
3212 case OFORMAT_HUMAN:
3213 if (e->data && !e->has_offset) {
3214 error_report("File contains external, encrypted or compressed clusters.");
3215 return -1;
3216 }
3217 if (e->data && !e->zero) {
3218 printf("%#-16"PRIx64"%#-16"PRIx64"%#-16"PRIx64"%s\n",
3219 e->start, e->length,
3220 e->has_offset ? e->offset : 0,
3221 e->filename ?: "");
3222 }
3223 /* This format ignores the distinction between 0, ZERO and ZERO|DATA.
3224 * Modify the flags here to allow more coalescing.
3225 */
3226 if (next && (!next->data || next->zero)) {
3227 next->data = false;
3228 next->zero = true;
3229 }
3230 break;
3231 case OFORMAT_JSON:
3232 printf("{ \"start\": %"PRId64", \"length\": %"PRId64","
3233 " \"depth\": %"PRId64", \"present\": %s, \"zero\": %s,"
3234 " \"data\": %s, \"compressed\": %s",
3235 e->start, e->length, e->depth,
3236 e->present ? "true" : "false",
3237 e->zero ? "true" : "false",
3238 e->data ? "true" : "false",
3239 e->compressed ? "true" : "false");
3240 if (e->has_offset) {
3241 printf(", \"offset\": %"PRId64"", e->offset);
3242 }
3243 putchar('}');
3244
3245 if (next) {
3246 puts(",");
3247 }
3248 break;
3249 }
3250 return 0;
3251 }
3252
3253 static int get_block_status(BlockDriverState *bs, int64_t offset,
3254 int64_t bytes, MapEntry *e)
3255 {
3256 int ret;
3257 int depth;
3258 BlockDriverState *file;
3259 bool has_offset;
3260 int64_t map;
3261 char *filename = NULL;
3262
3263 GLOBAL_STATE_CODE();
3264 GRAPH_RDLOCK_GUARD_MAINLOOP();
3265
3266 /* As an optimization, we could cache the current range of unallocated
3267 * clusters in each file of the chain, and avoid querying the same
3268 * range repeatedly.
3269 */
3270
3271 depth = 0;
3272 for (;;) {
3273 bs = bdrv_skip_filters(bs);
3274 ret = bdrv_block_status(bs, offset, bytes, &bytes, &map, &file);
3275 if (ret < 0) {
3276 return ret;
3277 }
3278 assert(bytes);
3279 if (ret & (BDRV_BLOCK_ZERO|BDRV_BLOCK_DATA)) {
3280 break;
3281 }
3282 bs = bdrv_cow_bs(bs);
3283 if (bs == NULL) {
3284 ret = 0;
3285 break;
3286 }
3287
3288 depth++;
3289 }
3290
3291 has_offset = !!(ret & BDRV_BLOCK_OFFSET_VALID);
3292
3293 if (file && has_offset) {
3294 bdrv_refresh_filename(file);
3295 filename = file->filename;
3296 }
3297
3298 *e = (MapEntry) {
3299 .start = offset,
3300 .length = bytes,
3301 .data = !!(ret & BDRV_BLOCK_DATA),
3302 .zero = !!(ret & BDRV_BLOCK_ZERO),
3303 .compressed = !!(ret & BDRV_BLOCK_COMPRESSED),
3304 .offset = map,
3305 .has_offset = has_offset,
3306 .depth = depth,
3307 .present = !!(ret & BDRV_BLOCK_ALLOCATED),
3308 .filename = filename,
3309 };
3310
3311 return 0;
3312 }
3313
3314 static inline bool entry_mergeable(const MapEntry *curr, const MapEntry *next)
3315 {
3316 if (curr->length == 0) {
3317 return false;
3318 }
3319 if (curr->zero != next->zero ||
3320 curr->data != next->data ||
3321 curr->compressed != next->compressed ||
3322 curr->depth != next->depth ||
3323 curr->present != next->present ||
3324 !curr->filename != !next->filename ||
3325 curr->has_offset != next->has_offset) {
3326 return false;
3327 }
3328 if (curr->filename && strcmp(curr->filename, next->filename)) {
3329 return false;
3330 }
3331 if (curr->has_offset && curr->offset + curr->length != next->offset) {
3332 return false;
3333 }
3334 return true;
3335 }
3336
3337 static int img_map(const img_cmd_t *ccmd, int argc, char **argv)
3338 {
3339 int c;
3340 OutputFormat output_format = OFORMAT_HUMAN;
3341 BlockBackend *blk;
3342 BlockDriverState *bs;
3343 const char *filename, *fmt;
3344 int64_t length;
3345 MapEntry curr = { .length = 0 }, next;
3346 int ret = 0;
3347 bool image_opts = false;
3348 bool force_share = false;
3349 int64_t start_offset = 0;
3350 int64_t max_length = -1;
3351
3352 fmt = NULL;
3353 for (;;) {
3354 static const struct option long_options[] = {
3355 {"help", no_argument, 0, 'h'},
3356 {"format", required_argument, 0, 'f'},
3357 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3358 {"start-offset", required_argument, 0, 's'},
3359 {"max-length", required_argument, 0, 'l'},
3360 {"force-share", no_argument, 0, 'U'},
3361 {"output", required_argument, 0, OPTION_OUTPUT},
3362 {"object", required_argument, 0, OPTION_OBJECT},
3363 {0, 0, 0, 0}
3364 };
3365 c = getopt_long(argc, argv, "hf:s:l:U",
3366 long_options, NULL);
3367 if (c == -1) {
3368 break;
3369 }
3370 switch (c) {
3371 case 'h':
3372 cmd_help(ccmd, "[-f FMT | --image-opts]\n"
3373 " [--start-offset OFFSET] [--max-length LENGTH]\n"
3374 " [--output human|json] [-U] [--object OBJDEF] FILE\n"
3375 ,
3376 " -f, --format FMT\n"
3377 " specify FILE image format explicitly (default: probing is used)\n"
3378 " --image-opts\n"
3379 " treat FILE as an option string (key=value,..), not a file name\n"
3380 " (incompatible with -f|--format)\n"
3381 " -s, --start-offset OFFSET\n"
3382 " start at the given OFFSET in the image, not at the beginning\n"
3383 " -l, --max-length LENGTH\n"
3384 " process at most LENGTH bytes instead of up to the end of the image\n"
3385 " --output human|json\n"
3386 " specify output format name (default: human)\n"
3387 " -U, --force-share\n"
3388 " open image in shared mode for concurrent access\n"
3389 " --object OBJDEF\n"
3390 " defines QEMU user-creatable object\n"
3391 " FILE\n"
3392 " the image file name, or option string (key=value,..)\n"
3393 " with --image-opts, to operate on\n"
3394 );
3395 break;
3396 case 'f':
3397 fmt = optarg;
3398 break;
3399 case OPTION_IMAGE_OPTS:
3400 image_opts = true;
3401 break;
3402 case 's':
3403 start_offset = cvtnum("start offset", optarg, true);
3404 if (start_offset < 0) {
3405 return 1;
3406 }
3407 break;
3408 case 'l':
3409 max_length = cvtnum("max length", optarg, true);
3410 if (max_length < 0) {
3411 return 1;
3412 }
3413 break;
3414 case OPTION_OUTPUT:
3415 output_format = parse_output_format(argv[0], optarg);
3416 break;
3417 case 'U':
3418 force_share = true;
3419 break;
3420 case OPTION_OBJECT:
3421 user_creatable_process_cmdline(optarg);
3422 break;
3423 default:
3424 tryhelp(argv[0]);
3425 }
3426 }
3427 if (optind != argc - 1) {
3428 error_exit(argv[0], "Expecting one image file name");
3429 }
3430 filename = argv[optind];
3431
3432 blk = img_open(image_opts, filename, fmt, 0, false, false, force_share);
3433 if (!blk) {
3434 return 1;
3435 }
3436 bs = blk_bs(blk);
3437
3438 if (output_format == OFORMAT_HUMAN) {
3439 printf("%-16s%-16s%-16s%s\n", "Offset", "Length", "Mapped to", "File");
3440 } else if (output_format == OFORMAT_JSON) {
3441 putchar('[');
3442 }
3443
3444 length = blk_getlength(blk);
3445 if (length < 0) {
3446 error_report("Failed to get size for '%s'", filename);
3447 return 1;
3448 }
3449 if (max_length != -1) {
3450 length = MIN(start_offset + max_length, length);
3451 }
3452
3453 curr.start = start_offset;
3454 while (curr.start + curr.length < length) {
3455 int64_t offset = curr.start + curr.length;
3456 int64_t n = length - offset;
3457
3458 ret = get_block_status(bs, offset, n, &next);
3459 if (ret < 0) {
3460 error_report("Could not read file metadata: %s", strerror(-ret));
3461 goto out;
3462 }
3463
3464 if (entry_mergeable(&curr, &next)) {
3465 curr.length += next.length;
3466 continue;
3467 }
3468
3469 if (curr.length > 0) {
3470 ret = dump_map_entry(output_format, &curr, &next);
3471 if (ret < 0) {
3472 goto out;
3473 }
3474 }
3475 curr = next;
3476 }
3477
3478 ret = dump_map_entry(output_format, &curr, NULL);
3479 if (output_format == OFORMAT_JSON) {
3480 puts("]");
3481 }
3482
3483 out:
3484 blk_unref(blk);
3485 return ret < 0;
3486 }
3487
3488 /* the same as options */
3489 #define SNAPSHOT_LIST 'l'
3490 #define SNAPSHOT_CREATE 'c'
3491 #define SNAPSHOT_APPLY 'a'
3492 #define SNAPSHOT_DELETE 'd'
3493
3494 static int img_snapshot(const img_cmd_t *ccmd, int argc, char **argv)
3495 {
3496 BlockBackend *blk;
3497 BlockDriverState *bs;
3498 QEMUSnapshotInfo sn;
3499 char *filename, *fmt = NULL, *snapshot_name = NULL;
3500 int c, ret = 0;
3501 int action = 0;
3502 bool quiet = false;
3503 Error *err = NULL;
3504 bool image_opts = false;
3505 bool force_share = false;
3506 int64_t rt;
3507
3508 /* Parse commandline parameters */
3509 for(;;) {
3510 static const struct option long_options[] = {
3511 {"help", no_argument, 0, 'h'},
3512 {"format", required_argument, 0, 'f'},
3513 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3514 {"list", no_argument, 0, SNAPSHOT_LIST},
3515 {"apply", required_argument, 0, SNAPSHOT_APPLY},
3516 {"create", required_argument, 0, SNAPSHOT_CREATE},
3517 {"delete", required_argument, 0, SNAPSHOT_DELETE},
3518 {"force-share", no_argument, 0, 'U'},
3519 {"quiet", no_argument, 0, 'q'},
3520 {"object", required_argument, 0, OPTION_OBJECT},
3521 {0, 0, 0, 0}
3522 };
3523 c = getopt_long(argc, argv, "hf:la:c:d:Uq",
3524 long_options, NULL);
3525 if (c == -1) {
3526 break;
3527 }
3528 switch(c) {
3529 case 'h':
3530 cmd_help(ccmd, "[-f FMT | --image-opts] [-l | -a|-c|-d SNAPSHOT]\n"
3531 " [-U] [-q] [--object OBJDEF] FILE\n"
3532 ,
3533 " -f, --format FMT\n"
3534 " specify FILE format explicitly (default: probing is used)\n"
3535 " --image-opts\n"
3536 " treat FILE as an option string (key=value,..), not a file name\n"
3537 " (incompatible with -f|--format)\n"
3538 " -l, --list\n"
3539 " list snapshots in FILE (default action if no -l|-c|-a|-d is given)\n"
3540 " -c, --create SNAPSHOT\n"
3541 " create named snapshot\n"
3542 " -a, --apply SNAPSHOT\n"
3543 " apply named snapshot to the base\n"
3544 " -d, --delete SNAPSHOT\n"
3545 " delete named snapshot\n"
3546 " (only one of -l|-c|-a|-d can be specified)\n"
3547 " -U, --force-share\n"
3548 " open image in shared mode for concurrent access\n"
3549 " -q, --quiet\n"
3550 " quiet mode (produce only error messages if any)\n"
3551 " --object OBJDEF\n"
3552 " defines QEMU user-creatable object\n"
3553 " FILE\n"
3554 " name of the image file, or option string (key=value,..)\n"
3555 " with --image-opts) to operate on\n"
3556 );
3557 break;
3558 case 'f':
3559 fmt = optarg;
3560 break;
3561 case OPTION_IMAGE_OPTS:
3562 image_opts = true;
3563 break;
3564 case SNAPSHOT_LIST:
3565 case SNAPSHOT_APPLY:
3566 case SNAPSHOT_CREATE:
3567 case SNAPSHOT_DELETE:
3568 if (action) {
3569 error_exit(argv[0], "Cannot mix '-l', '-a', '-c', '-d'");
3570 return 0;
3571 }
3572 action = c;
3573 snapshot_name = optarg;
3574 break;
3575 case 'U':
3576 force_share = true;
3577 break;
3578 case 'q':
3579 quiet = true;
3580 break;
3581 case OPTION_OBJECT:
3582 user_creatable_process_cmdline(optarg);
3583 break;
3584 default:
3585 tryhelp(argv[0]);
3586 }
3587 }
3588
3589 if (optind != argc - 1) {
3590 error_exit(argv[0], "Expecting one image file name");
3591 }
3592 filename = argv[optind++];
3593
3594 if (!action) {
3595 action = SNAPSHOT_LIST;
3596 }
3597
3598 /* Open the image */
3599 blk = img_open(image_opts, filename, fmt,
3600 action == SNAPSHOT_LIST ? 0 : BDRV_O_RDWR,
3601 false, quiet, force_share);
3602 if (!blk) {
3603 return 1;
3604 }
3605 bs = blk_bs(blk);
3606
3607 /* Perform the requested action */
3608 switch(action) {
3609 case SNAPSHOT_LIST:
3610 dump_snapshots(bs);
3611 break;
3612
3613 case SNAPSHOT_CREATE:
3614 memset(&sn, 0, sizeof(sn));
3615 pstrcpy(sn.name, sizeof(sn.name), snapshot_name);
3616
3617 rt = g_get_real_time();
3618 sn.date_sec = rt / G_USEC_PER_SEC;
3619 sn.date_nsec = (rt % G_USEC_PER_SEC) * 1000;
3620
3621 bdrv_graph_rdlock_main_loop();
3622 ret = bdrv_snapshot_create(bs, &sn);
3623 bdrv_graph_rdunlock_main_loop();
3624
3625 if (ret) {
3626 error_report("Could not create snapshot '%s': %s",
3627 snapshot_name, strerror(-ret));
3628 }
3629 break;
3630
3631 case SNAPSHOT_APPLY:
3632 ret = bdrv_snapshot_goto(bs, snapshot_name, &err);
3633 if (ret) {
3634 error_reportf_err(err, "Could not apply snapshot '%s': ",
3635 snapshot_name);
3636 }
3637 break;
3638
3639 case SNAPSHOT_DELETE:
3640 bdrv_drain_all_begin();
3641 bdrv_graph_rdlock_main_loop();
3642 ret = bdrv_snapshot_find(bs, &sn, snapshot_name);
3643 if (ret < 0) {
3644 error_report("Could not delete snapshot '%s': snapshot not "
3645 "found", snapshot_name);
3646 ret = 1;
3647 } else {
3648 ret = bdrv_snapshot_delete(bs, sn.id_str, sn.name, &err);
3649 if (ret < 0) {
3650 error_reportf_err(err, "Could not delete snapshot '%s': ",
3651 snapshot_name);
3652 ret = 1;
3653 }
3654 }
3655 bdrv_graph_rdunlock_main_loop();
3656 bdrv_drain_all_end();
3657 break;
3658 }
3659
3660 /* Cleanup */
3661 blk_unref(blk);
3662 if (ret) {
3663 return 1;
3664 }
3665 return 0;
3666 }
3667
3668 static int img_rebase(const img_cmd_t *ccmd, int argc, char **argv)
3669 {
3670 BlockBackend *blk = NULL, *blk_old_backing = NULL, *blk_new_backing = NULL;
3671 uint8_t *buf_old = NULL;
3672 uint8_t *buf_new = NULL;
3673 BlockDriverState *bs = NULL, *prefix_chain_bs = NULL;
3674 BlockDriverState *unfiltered_bs, *unfiltered_bs_cow;
3675 BlockDriverInfo bdi = {0};
3676 char *filename;
3677 const char *fmt, *cache, *src_cache, *out_basefmt, *out_baseimg;
3678 int c, flags, src_flags, ret;
3679 BdrvRequestFlags write_flags = 0;
3680 bool writethrough, src_writethrough;
3681 int unsafe = 0;
3682 bool force_share = false;
3683 int progress = 0;
3684 bool quiet = false;
3685 bool compress = false;
3686 Error *local_err = NULL;
3687 bool image_opts = false;
3688 int64_t write_align;
3689
3690 /* Parse commandline parameters */
3691 fmt = NULL;
3692 cache = BDRV_DEFAULT_CACHE;
3693 src_cache = BDRV_DEFAULT_CACHE;
3694 out_baseimg = NULL;
3695 out_basefmt = NULL;
3696 for(;;) {
3697 static const struct option long_options[] = {
3698 {"help", no_argument, 0, 'h'},
3699 {"format", required_argument, 0, 'f'},
3700 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
3701 {"cache", required_argument, 0, 't'},
3702 {"compress", no_argument, 0, 'c'},
3703 {"backing", required_argument, 0, 'b'},
3704 {"backing-format", required_argument, 0, 'B'},
3705 {"backing-cache", required_argument, 0, 'T'},
3706 {"backing-unsafe", no_argument, 0, 'u'},
3707 {"force-share", no_argument, 0, 'U'},
3708 {"progress", no_argument, 0, 'p'},
3709 {"quiet", no_argument, 0, 'q'},
3710 {"object", required_argument, 0, OPTION_OBJECT},
3711 {0, 0, 0, 0}
3712 };
3713 c = getopt_long(argc, argv, "hf:t:cb:F:B:T:uUpq",
3714 long_options, NULL);
3715 if (c == -1) {
3716 break;
3717 }
3718 switch (c) {
3719 case 'h':
3720 cmd_help(ccmd, "[-f FMT | --image-opts] [-t CACHE]\n"
3721 " [-b BACKING_FILE [-B BACKING_FMT] [-T BACKING_CACHE]] [-u]\n"
3722 " [-c] [-U] [-p] [-q] [--object OBJDEF] FILE\n"
3723 ,
3724 " -f, --format FMT\n"
3725 " specify FILE format explicitly (default: probing is used)\n"
3726 " --image-opts\n"
3727 " treat FILE as an option string (key=value,..), not a file name\n"
3728 " (incompatible with -f|--format)\n"
3729 " -t, --cache CACHE\n"
3730 " cache mode for FILE (default: " BDRV_DEFAULT_CACHE ")\n"
3731 " -b, --backing BACKING_FILE|\"\"\n"
3732 " rebase onto this file (specify empty name for no backing file)\n"
3733 " -B, --backing-format BACKING_FMT (was -F in <=10.0)\n"
3734 " specify format for BACKING_FILE explicitly (default: probing is used)\n"
3735 " -T, --backing-cache CACHE\n"
3736 " BACKING_FILE cache mode (default: " BDRV_DEFAULT_CACHE ")\n"
3737 " -u, --backing-unsafe\n"
3738 " do not fail if BACKING_FILE can not be read\n"
3739 " -c, --compress\n"
3740 " compress image (when image supports this)\n"
3741 " -U, --force-share\n"
3742 " open image in shared mode for concurrent access\n"
3743 " -p, --progress\n"
3744 " display progress information\n"
3745 " -q, --quiet\n"
3746 " quiet mode (produce only error messages if any)\n"
3747 " --object OBJDEF\n"
3748 " defines QEMU user-creatable object\n"
3749 " FILE\n"
3750 " name of the image file, or option string (key=value,..)\n"
3751 " with --image-opts, to operate on\n"
3752 );
3753 return 0;
3754 case 'f':
3755 fmt = optarg;
3756 break;
3757 case OPTION_IMAGE_OPTS:
3758 image_opts = true;
3759 break;
3760 case 't':
3761 cache = optarg;
3762 break;
3763 case 'b':
3764 out_baseimg = optarg;
3765 break;
3766 case 'F': /* <=10.0 */
3767 case 'B':
3768 out_basefmt = optarg;
3769 break;
3770 case 'u':
3771 unsafe = 1;
3772 break;
3773 case 'c':
3774 compress = true;
3775 break;
3776 case 'U':
3777 force_share = true;
3778 break;
3779 case 'p':
3780 progress = 1;
3781 break;
3782 case 'T':
3783 src_cache = optarg;
3784 break;
3785 case 'q':
3786 quiet = true;
3787 break;
3788 case OPTION_OBJECT:
3789 user_creatable_process_cmdline(optarg);
3790 break;
3791 default:
3792 tryhelp(argv[0]);
3793 }
3794 }
3795
3796 if (quiet) {
3797 progress = 0;
3798 }
3799
3800 if (optind != argc - 1) {
3801 error_exit(argv[0], "Expecting one image file name");
3802 }
3803 if (!unsafe && !out_baseimg) {
3804 error_exit(argv[0],
3805 "Must specify backing file (-b) or use unsafe mode (-u)");
3806 }
3807 filename = argv[optind++];
3808
3809 qemu_progress_init(progress, 2.0);
3810 qemu_progress_print(0, 100);
3811
3812 flags = BDRV_O_RDWR | (unsafe ? BDRV_O_NO_BACKING : 0);
3813 ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
3814 if (ret < 0) {
3815 error_report("Invalid cache option: %s", cache);
3816 goto out;
3817 }
3818
3819 src_flags = 0;
3820 ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
3821 if (ret < 0) {
3822 error_report("Invalid source cache option: %s", src_cache);
3823 goto out;
3824 }
3825
3826 /* The source files are opened read-only, don't care about WCE */
3827 assert((src_flags & BDRV_O_RDWR) == 0);
3828 (void) src_writethrough;
3829
3830 /*
3831 * Open the images.
3832 *
3833 * Ignore the old backing file for unsafe rebase in case we want to correct
3834 * the reference to a renamed or moved backing file.
3835 */
3836 blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
3837 false);
3838 if (!blk) {
3839 ret = -1;
3840 goto out;
3841 }
3842 bs = blk_bs(blk);
3843
3844 bdrv_graph_rdlock_main_loop();
3845 unfiltered_bs = bdrv_skip_filters(bs);
3846 unfiltered_bs_cow = bdrv_cow_bs(unfiltered_bs);
3847 bdrv_graph_rdunlock_main_loop();
3848
3849 if (compress && !block_driver_can_compress(unfiltered_bs->drv)) {
3850 error_report("Compression not supported for this file format");
3851 ret = -1;
3852 goto out;
3853 } else if (compress) {
3854 write_flags |= BDRV_REQ_WRITE_COMPRESSED;
3855 }
3856
3857 if (out_basefmt != NULL) {
3858 if (bdrv_find_format(out_basefmt) == NULL) {
3859 error_report("Invalid format name: '%s'", out_basefmt);
3860 ret = -1;
3861 goto out;
3862 }
3863 }
3864
3865 /*
3866 * We need overlay subcluster size (or cluster size in case writes are
3867 * compressed) to make sure write requests are aligned.
3868 */
3869 ret = bdrv_get_info(unfiltered_bs, &bdi);
3870 if (ret < 0) {
3871 error_report("could not get block driver info");
3872 goto out;
3873 } else if (bdi.subcluster_size == 0) {
3874 bdi.cluster_size = bdi.subcluster_size = 1;
3875 }
3876
3877 write_align = compress ? bdi.cluster_size : bdi.subcluster_size;
3878
3879 /* For safe rebasing we need to compare old and new backing file */
3880 if (!unsafe) {
3881 QDict *options = NULL;
3882 BlockDriverState *base_bs;
3883
3884 bdrv_graph_rdlock_main_loop();
3885 base_bs = bdrv_cow_bs(unfiltered_bs);
3886 bdrv_graph_rdunlock_main_loop();
3887
3888 if (base_bs) {
3889 blk_old_backing = blk_new(qemu_get_aio_context(),
3890 BLK_PERM_CONSISTENT_READ,
3891 BLK_PERM_ALL);
3892 ret = blk_insert_bs(blk_old_backing, base_bs,
3893 &local_err);
3894 if (ret < 0) {
3895 error_reportf_err(local_err,
3896 "Could not reuse old backing file '%s': ",
3897 base_bs->filename);
3898 goto out;
3899 }
3900 } else {
3901 blk_old_backing = NULL;
3902 }
3903
3904 if (out_baseimg[0]) {
3905 const char *overlay_filename;
3906 char *out_real_path;
3907
3908 options = qdict_new();
3909 if (out_basefmt) {
3910 qdict_put_str(options, "driver", out_basefmt);
3911 }
3912 if (force_share) {
3913 qdict_put_bool(options, BDRV_OPT_FORCE_SHARE, true);
3914 }
3915
3916 bdrv_graph_rdlock_main_loop();
3917 bdrv_refresh_filename(bs);
3918 bdrv_graph_rdunlock_main_loop();
3919 overlay_filename = bs->exact_filename[0] ? bs->exact_filename
3920 : bs->filename;
3921 out_real_path =
3922 bdrv_get_full_backing_filename_from_filename(overlay_filename,
3923 out_baseimg,
3924 &local_err);
3925 if (local_err) {
3926 qobject_unref(options);
3927 error_reportf_err(local_err,
3928 "Could not resolve backing filename: ");
3929 ret = -1;
3930 goto out;
3931 }
3932
3933 /*
3934 * Find out whether we rebase an image on top of a previous image
3935 * in its chain.
3936 */
3937 prefix_chain_bs = bdrv_find_backing_image(bs, out_real_path);
3938 if (prefix_chain_bs) {
3939 qobject_unref(options);
3940 g_free(out_real_path);
3941
3942 blk_new_backing = blk_new(qemu_get_aio_context(),
3943 BLK_PERM_CONSISTENT_READ,
3944 BLK_PERM_ALL);
3945 ret = blk_insert_bs(blk_new_backing, prefix_chain_bs,
3946 &local_err);
3947 if (ret < 0) {
3948 error_reportf_err(local_err,
3949 "Could not reuse backing file '%s': ",
3950 out_baseimg);
3951 goto out;
3952 }
3953 } else {
3954 blk_new_backing = blk_new_open(out_real_path, NULL,
3955 options, src_flags, &local_err);
3956 g_free(out_real_path);
3957 if (!blk_new_backing) {
3958 error_reportf_err(local_err,
3959 "Could not open new backing file '%s': ",
3960 out_baseimg);
3961 ret = -1;
3962 goto out;
3963 }
3964 }
3965 }
3966 }
3967
3968 /*
3969 * Check each unallocated cluster in the COW file. If it is unallocated,
3970 * accesses go to the backing file. We must therefore compare this cluster
3971 * in the old and new backing file, and if they differ we need to copy it
3972 * from the old backing file into the COW file.
3973 *
3974 * If qemu-img crashes during this step, no harm is done. The content of
3975 * the image is the same as the original one at any time.
3976 */
3977 if (!unsafe) {
3978 int64_t size;
3979 int64_t old_backing_size = 0;
3980 int64_t new_backing_size = 0;
3981 uint64_t offset;
3982 int64_t n, n_old = 0, n_new = 0;
3983 float local_progress = 0;
3984
3985 if (blk_old_backing && bdrv_opt_mem_align(blk_bs(blk_old_backing)) >
3986 bdrv_opt_mem_align(blk_bs(blk))) {
3987 buf_old = blk_blockalign(blk_old_backing, IO_BUF_SIZE);
3988 } else {
3989 buf_old = blk_blockalign(blk, IO_BUF_SIZE);
3990 }
3991 buf_new = blk_blockalign(blk_new_backing, IO_BUF_SIZE);
3992
3993 size = blk_getlength(blk);
3994 if (size < 0) {
3995 error_report("Could not get size of '%s': %s",
3996 filename, strerror(-size));
3997 ret = -1;
3998 goto out;
3999 }
4000 if (blk_old_backing) {
4001 old_backing_size = blk_getlength(blk_old_backing);
4002 if (old_backing_size < 0) {
4003 char backing_name[PATH_MAX];
4004
4005 bdrv_get_backing_filename(bs, backing_name,
4006 sizeof(backing_name));
4007 error_report("Could not get size of '%s': %s",
4008 backing_name, strerror(-old_backing_size));
4009 ret = -1;
4010 goto out;
4011 }
4012 }
4013 if (blk_new_backing) {
4014 new_backing_size = blk_getlength(blk_new_backing);
4015 if (new_backing_size < 0) {
4016 error_report("Could not get size of '%s': %s",
4017 out_baseimg, strerror(-new_backing_size));
4018 ret = -1;
4019 goto out;
4020 }
4021 }
4022
4023 if (size != 0) {
4024 local_progress = (float)100 / (size / MIN(size, IO_BUF_SIZE));
4025 }
4026
4027 for (offset = 0; offset < size; offset += n) {
4028 bool old_backing_eof = false;
4029 int64_t n_alloc;
4030
4031 /* How many bytes can we handle with the next read? */
4032 n = MIN(IO_BUF_SIZE, size - offset);
4033
4034 /* If the cluster is allocated, we don't need to take action */
4035 ret = bdrv_is_allocated(unfiltered_bs, offset, n, &n);
4036 if (ret < 0) {
4037 error_report("error while reading image metadata: %s",
4038 strerror(-ret));
4039 goto out;
4040 }
4041 if (ret) {
4042 continue;
4043 }
4044
4045 if (prefix_chain_bs) {
4046 uint64_t bytes = n;
4047
4048 /*
4049 * If cluster wasn't changed since prefix_chain, we don't need
4050 * to take action
4051 */
4052 ret = bdrv_is_allocated_above(unfiltered_bs_cow,
4053 prefix_chain_bs, false,
4054 offset, n, &n);
4055 if (ret < 0) {
4056 error_report("error while reading image metadata: %s",
4057 strerror(-ret));
4058 goto out;
4059 }
4060 if (!ret && n) {
4061 continue;
4062 }
4063 if (!n) {
4064 /*
4065 * If we've reached EOF of the old backing, it means that
4066 * offsets beyond the old backing size were read as zeroes.
4067 * Now we will need to explicitly zero the cluster in
4068 * order to preserve that state after the rebase.
4069 */
4070 n = bytes;
4071 }
4072 }
4073
4074 /*
4075 * At this point we know that the region [offset; offset + n)
4076 * is unallocated within the target image. This region might be
4077 * unaligned to the target image's (sub)cluster boundaries, as
4078 * old backing may have smaller clusters (or have subclusters).
4079 * We extend it to the aligned boundaries to avoid CoW on
4080 * partial writes in blk_pwrite(),
4081 */
4082 n += offset - QEMU_ALIGN_DOWN(offset, write_align);
4083 offset = QEMU_ALIGN_DOWN(offset, write_align);
4084 n += QEMU_ALIGN_UP(offset + n, write_align) - (offset + n);
4085 n = MIN(n, MIN(size - offset, IO_BUF_SIZE));
4086 assert(!bdrv_is_allocated(unfiltered_bs, offset, n, &n_alloc) &&
4087 n_alloc == n);
4088
4089 /*
4090 * Much like with the target image, we'll try to read as much
4091 * of the old and new backings as we can.
4092 */
4093 n_old = MIN(n, MAX(0, old_backing_size - (int64_t) offset));
4094 n_new = MIN(n, MAX(0, new_backing_size - (int64_t) offset));
4095
4096 /*
4097 * Read old and new backing file and take into consideration that
4098 * backing files may be smaller than the COW image.
4099 */
4100 memset(buf_old + n_old, 0, n - n_old);
4101 if (!n_old) {
4102 old_backing_eof = true;
4103 } else {
4104 ret = blk_pread(blk_old_backing, offset, n_old, buf_old, 0);
4105 if (ret < 0) {
4106 error_report("error while reading from old backing file");
4107 goto out;
4108 }
4109 }
4110
4111 memset(buf_new + n_new, 0, n - n_new);
4112 if (n_new) {
4113 ret = blk_pread(blk_new_backing, offset, n_new, buf_new, 0);
4114 if (ret < 0) {
4115 error_report("error while reading from new backing file");
4116 goto out;
4117 }
4118 }
4119
4120 /* If they differ, we need to write to the COW file */
4121 uint64_t written = 0;
4122
4123 while (written < n) {
4124 int64_t pnum;
4125
4126 if (compare_buffers(buf_old + written, buf_new + written,
4127 n - written, write_align, &pnum))
4128 {
4129 if (old_backing_eof) {
4130 ret = blk_pwrite_zeroes(blk, offset + written, pnum, 0);
4131 } else {
4132 assert(written + pnum <= IO_BUF_SIZE);
4133 ret = blk_pwrite(blk, offset + written, pnum,
4134 buf_old + written, write_flags);
4135 }
4136 if (ret < 0) {
4137 error_report("Error while writing to COW image: %s",
4138 strerror(-ret));
4139 goto out;
4140 }
4141 }
4142
4143 written += pnum;
4144 if (offset + written >= old_backing_size) {
4145 old_backing_eof = true;
4146 }
4147 }
4148 qemu_progress_print(local_progress, 100);
4149 }
4150 }
4151
4152 /*
4153 * Change the backing file. All clusters that are different from the old
4154 * backing file are overwritten in the COW file now, so the visible content
4155 * doesn't change when we switch the backing file.
4156 */
4157 if (out_baseimg && *out_baseimg) {
4158 ret = bdrv_change_backing_file(unfiltered_bs, out_baseimg, out_basefmt,
4159 true);
4160 } else {
4161 ret = bdrv_change_backing_file(unfiltered_bs, NULL, NULL, false);
4162 }
4163
4164 if (ret == -ENOSPC) {
4165 error_report("Could not change the backing file to '%s': No "
4166 "space left in the file header", out_baseimg);
4167 } else if (ret == -EINVAL && out_baseimg && !out_basefmt) {
4168 error_report("Could not change the backing file to '%s': backing "
4169 "format must be specified", out_baseimg);
4170 } else if (ret < 0) {
4171 error_report("Could not change the backing file to '%s': %s",
4172 out_baseimg, strerror(-ret));
4173 }
4174
4175 qemu_progress_print(100, 0);
4176 /*
4177 * TODO At this point it is possible to check if any clusters that are
4178 * allocated in the COW file are the same in the backing file. If so, they
4179 * could be dropped from the COW file. Don't do this before switching the
4180 * backing file, in case of a crash this would lead to corruption.
4181 */
4182 out:
4183 qemu_progress_end();
4184 /* Cleanup */
4185 if (!unsafe) {
4186 blk_unref(blk_old_backing);
4187 blk_unref(blk_new_backing);
4188 }
4189 qemu_vfree(buf_old);
4190 qemu_vfree(buf_new);
4191
4192 blk_unref(blk);
4193 if (ret) {
4194 return 1;
4195 }
4196 return 0;
4197 }
4198
4199 static int img_resize(const img_cmd_t *ccmd, int argc, char **argv)
4200 {
4201 Error *err = NULL;
4202 int c, ret, relative;
4203 const char *filename = NULL, *fmt = NULL, *size = NULL;
4204 int64_t n, total_size, current_size;
4205 bool quiet = false;
4206 BlockBackend *blk = NULL;
4207 PreallocMode prealloc = PREALLOC_MODE_OFF;
4208 QemuOpts *param;
4209
4210 static QemuOptsList resize_options = {
4211 .name = "resize_options",
4212 .head = QTAILQ_HEAD_INITIALIZER(resize_options.head),
4213 .desc = {
4214 {
4215 .name = BLOCK_OPT_SIZE,
4216 .type = QEMU_OPT_SIZE,
4217 .help = "Virtual disk size"
4218 }, {
4219 /* end of list */
4220 }
4221 },
4222 };
4223 bool image_opts = false;
4224 bool shrink = false;
4225
4226 /* Parse getopt arguments */
4227 for(;;) {
4228 static const struct option long_options[] = {
4229 {"help", no_argument, 0, 'h'},
4230 {"format", required_argument, 0, 'f'},
4231 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4232 {"preallocation", required_argument, 0, OPTION_PREALLOCATION},
4233 {"shrink", no_argument, 0, OPTION_SHRINK},
4234 {"quiet", no_argument, 0, 'q'},
4235 {"object", required_argument, 0, OPTION_OBJECT},
4236 {0, 0, 0, 0}
4237 };
4238 c = getopt_long(argc, argv, "-hf:q",
4239 long_options, NULL);
4240 if (c == -1) {
4241 break;
4242 }
4243 switch(c) {
4244 case 'h':
4245 cmd_help(ccmd, "[-f FMT | --image-opts] [--preallocation PREALLOC] [--shrink]\n"
4246 " [-q] [--object OBJDEF] FILE [+-]SIZE[bkKMGTPE]\n"
4247 ,
4248 " -f, --format FMT\n"
4249 " specify FILE format explicitly (default: probing is used)\n"
4250 " --image-opts\n"
4251 " treat FILE as an option string (key=value,...), not a file name\n"
4252 " (incompatible with -f|--format)\n"
4253 " --shrink\n"
4254 " allow operation when the new size is smaller than the original\n"
4255 " --preallocation PREALLOC\n"
4256 " specify FMT-specific preallocation type for the new areas\n"
4257 " -q, --quiet\n"
4258 " quiet mode (produce only error messages if any)\n"
4259 " --object OBJDEF\n"
4260 " defines QEMU user-creatable object\n"
4261 " FILE\n"
4262 " name of the image file, or option string (key=value,..)\n"
4263 " with --image-opts, to operate on\n"
4264 " [+-]SIZE[bkKMGTPE]\n"
4265 " new image size or amount by which to shrink (-)/grow (+),\n"
4266 " with optional multiplier suffix (powers of 1024, default is bytes)\n"
4267 );
4268 return 0;
4269 case 'f':
4270 fmt = optarg;
4271 break;
4272 case OPTION_IMAGE_OPTS:
4273 image_opts = true;
4274 break;
4275 case OPTION_PREALLOCATION:
4276 prealloc = qapi_enum_parse(&PreallocMode_lookup, optarg,
4277 PREALLOC_MODE__MAX, NULL);
4278 if (prealloc == PREALLOC_MODE__MAX) {
4279 error_report("Invalid preallocation mode '%s'", optarg);
4280 return 1;
4281 }
4282 break;
4283 case OPTION_SHRINK:
4284 shrink = true;
4285 break;
4286 case 'q':
4287 quiet = true;
4288 break;
4289 case OPTION_OBJECT:
4290 user_creatable_process_cmdline(optarg);
4291 break;
4292 case 1: /* a non-optional argument */
4293 if (!filename) {
4294 filename = optarg;
4295 /* see if we have -size (number) next to filename */
4296 if (optind < argc) {
4297 size = argv[optind];
4298 if (size[0] == '-' && size[1] >= '0' && size[1] <= '9') {
4299 ++optind;
4300 } else {
4301 size = NULL;
4302 }
4303 }
4304 } else if (!size) {
4305 size = optarg;
4306 } else {
4307 error_exit(argv[0], "Extra argument(s) in command line");
4308 }
4309 break;
4310 default:
4311 tryhelp(argv[0]);
4312 }
4313 }
4314 if (!filename && optind < argc) {
4315 filename = argv[optind++];
4316 }
4317 if (!size && optind < argc) {
4318 size = argv[optind++];
4319 }
4320 if (!filename || !size || optind < argc) {
4321 error_exit(argv[0], "Expecting image file name and size");
4322 }
4323
4324 /* Choose grow, shrink, or absolute resize mode */
4325 switch (size[0]) {
4326 case '+':
4327 relative = 1;
4328 size++;
4329 break;
4330 case '-':
4331 relative = -1;
4332 size++;
4333 break;
4334 default:
4335 relative = 0;
4336 break;
4337 }
4338
4339 /* Parse size */
4340 param = qemu_opts_create(&resize_options, NULL, 0, &error_abort);
4341 if (!qemu_opt_set(param, BLOCK_OPT_SIZE, size, &err)) {
4342 error_report_err(err);
4343 ret = -1;
4344 qemu_opts_del(param);
4345 goto out;
4346 }
4347 n = qemu_opt_get_size(param, BLOCK_OPT_SIZE, 0);
4348 qemu_opts_del(param);
4349
4350 blk = img_open(image_opts, filename, fmt,
4351 BDRV_O_RDWR | BDRV_O_RESIZE, false, quiet,
4352 false);
4353 if (!blk) {
4354 ret = -1;
4355 goto out;
4356 }
4357
4358 current_size = blk_getlength(blk);
4359 if (current_size < 0) {
4360 error_report("Failed to inquire current image length: %s",
4361 strerror(-current_size));
4362 ret = -1;
4363 goto out;
4364 }
4365
4366 if (relative) {
4367 total_size = current_size + n * relative;
4368 } else {
4369 total_size = n;
4370 }
4371 if (total_size <= 0) {
4372 error_report("New image size must be positive");
4373 ret = -1;
4374 goto out;
4375 }
4376
4377 if (total_size <= current_size && prealloc != PREALLOC_MODE_OFF) {
4378 error_report("Preallocation can only be used for growing images");
4379 ret = -1;
4380 goto out;
4381 }
4382
4383 if (total_size < current_size && !shrink) {
4384 error_report("Use the --shrink option to perform a shrink operation.");
4385 warn_report("Shrinking an image will delete all data beyond the "
4386 "shrunken image's end. Before performing such an "
4387 "operation, make sure there is no important data there.");
4388 ret = -1;
4389 goto out;
4390 }
4391
4392 /*
4393 * The user expects the image to have the desired size after
4394 * resizing, so pass @exact=true. It is of no use to report
4395 * success when the image has not actually been resized.
4396 */
4397 ret = blk_truncate(blk, total_size, true, prealloc, 0, &err);
4398 if (!ret) {
4399 qprintf(quiet, "Image resized.\n");
4400 } else {
4401 error_report_err(err);
4402 }
4403 out:
4404 blk_unref(blk);
4405 if (ret) {
4406 return 1;
4407 }
4408 return 0;
4409 }
4410
4411 static void amend_status_cb(BlockDriverState *bs,
4412 int64_t offset, int64_t total_work_size,
4413 void *opaque)
4414 {
4415 qemu_progress_print(100.f * offset / total_work_size, 0);
4416 }
4417
4418 static int print_amend_option_help(const char *format)
4419 {
4420 BlockDriver *drv;
4421
4422 GRAPH_RDLOCK_GUARD_MAINLOOP();
4423
4424 /* Find driver and parse its options */
4425 drv = bdrv_find_format(format);
4426 if (!drv) {
4427 error_report("Unknown file format '%s'", format);
4428 return 1;
4429 }
4430
4431 if (!drv->bdrv_amend_options) {
4432 error_report("Format driver '%s' does not support option amendment",
4433 format);
4434 return 1;
4435 }
4436
4437 /* Every driver supporting amendment must have amend_opts */
4438 assert(drv->amend_opts);
4439
4440 printf("Amend options for '%s':\n", format);
4441 qemu_opts_print_help(drv->amend_opts, false);
4442 return 0;
4443 }
4444
4445 static int img_amend(const img_cmd_t *ccmd, int argc, char **argv)
4446 {
4447 Error *err = NULL;
4448 int c, ret = 0;
4449 char *options = NULL;
4450 QemuOptsList *amend_opts = NULL;
4451 QemuOpts *opts = NULL;
4452 const char *fmt = NULL, *filename, *cache;
4453 int flags;
4454 bool writethrough;
4455 bool quiet = false, progress = false;
4456 BlockBackend *blk = NULL;
4457 BlockDriverState *bs = NULL;
4458 bool image_opts = false;
4459 bool force = false;
4460
4461 cache = BDRV_DEFAULT_CACHE;
4462 for (;;) {
4463 static const struct option long_options[] = {
4464 {"help", no_argument, 0, 'h'},
4465 {"options", required_argument, 0, 'o'},
4466 {"format", required_argument, 0, 'f'},
4467 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4468 {"cache", required_argument, 0, 't'},
4469 {"force", no_argument, 0, OPTION_FORCE},
4470 {"progress", no_argument, 0, 'p'},
4471 {"quiet", no_argument, 0, 'q'},
4472 {"object", required_argument, 0, OPTION_OBJECT},
4473 {0, 0, 0, 0}
4474 };
4475 c = getopt_long(argc, argv, "ho:f:t:pq",
4476 long_options, NULL);
4477 if (c == -1) {
4478 break;
4479 }
4480
4481 switch (c) {
4482 case 'h':
4483 cmd_help(ccmd, "-o FMT_OPTS [-f FMT | --image-opts]\n"
4484 " [-t CACHE] [--force] [-p] [-q] [--object OBJDEF] FILE\n"
4485 ,
4486 " -o, --options FMT_OPTS\n"
4487 " FMT-specfic format options (required)\n"
4488 " -f, --format FMT\n"
4489 " specify FILE format explicitly (default: probing is used)\n"
4490 " --image-opts\n"
4491 " treat FILE as an option string (key=value,..), not a file name\n"
4492 " (incompatible with -f|--format)\n"
4493 " -t, --cache CACHE\n"
4494 " cache mode for FILE (default: " BDRV_DEFAULT_CACHE ")\n"
4495 " --force\n"
4496 " allow certain unsafe operations\n"
4497 " -p, --progres\n"
4498 " show operation progress\n"
4499 " -q, --quiet\n"
4500 " quiet mode (produce only error messages if any)\n"
4501 " --object OBJDEF\n"
4502 " defines QEMU user-creatable object\n"
4503 " FILE\n"
4504 " name of the image file, or option string (key=value,..)\n"
4505 " with --image-opts, to operate on\n"
4506 );
4507 break;
4508 case 'o':
4509 if (accumulate_options(&options, optarg) < 0) {
4510 ret = -1;
4511 goto out_no_progress;
4512 }
4513 break;
4514 case 'f':
4515 fmt = optarg;
4516 break;
4517 case OPTION_IMAGE_OPTS:
4518 image_opts = true;
4519 break;
4520 case 't':
4521 cache = optarg;
4522 break;
4523 case OPTION_FORCE:
4524 force = true;
4525 break;
4526 case 'p':
4527 progress = true;
4528 break;
4529 case 'q':
4530 quiet = true;
4531 break;
4532 case OPTION_OBJECT:
4533 user_creatable_process_cmdline(optarg);
4534 break;
4535 default:
4536 tryhelp(argv[0]);
4537 }
4538 }
4539
4540 if (!options) {
4541 error_exit(argv[0], "Must specify options (-o)");
4542 }
4543
4544 if (quiet) {
4545 progress = false;
4546 }
4547 qemu_progress_init(progress, 1.0);
4548
4549 filename = (optind == argc - 1) ? argv[argc - 1] : NULL;
4550 if (fmt && has_help_option(options)) {
4551 /* If a format is explicitly specified (and possibly no filename is
4552 * given), print option help here */
4553 ret = print_amend_option_help(fmt);
4554 goto out;
4555 }
4556
4557 if (optind != argc - 1) {
4558 error_report("Expecting one image file name");
4559 ret = -1;
4560 goto out;
4561 }
4562
4563 flags = BDRV_O_RDWR;
4564 ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
4565 if (ret < 0) {
4566 error_report("Invalid cache option: %s", cache);
4567 goto out;
4568 }
4569
4570 blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
4571 false);
4572 if (!blk) {
4573 ret = -1;
4574 goto out;
4575 }
4576 bs = blk_bs(blk);
4577
4578 fmt = bs->drv->format_name;
4579
4580 if (has_help_option(options)) {
4581 /* If the format was auto-detected, print option help here */
4582 ret = print_amend_option_help(fmt);
4583 goto out;
4584 }
4585
4586 bdrv_graph_rdlock_main_loop();
4587 if (!bs->drv->bdrv_amend_options) {
4588 error_report("Format driver '%s' does not support option amendment",
4589 fmt);
4590 bdrv_graph_rdunlock_main_loop();
4591 ret = -1;
4592 goto out;
4593 }
4594
4595 /* Every driver supporting amendment must have amend_opts */
4596 assert(bs->drv->amend_opts);
4597
4598 amend_opts = qemu_opts_append(amend_opts, bs->drv->amend_opts);
4599 opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
4600 if (!qemu_opts_do_parse(opts, options, NULL, &err)) {
4601 qemu_opts_del(opts);
4602 /* Try to parse options using the create options */
4603 amend_opts = qemu_opts_append(amend_opts, bs->drv->create_opts);
4604 opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
4605 if (qemu_opts_do_parse(opts, options, NULL, NULL)) {
4606 error_append_hint(&err,
4607 "This option is only supported for image creation\n");
4608 }
4609
4610 bdrv_graph_rdunlock_main_loop();
4611 error_report_err(err);
4612 ret = -1;
4613 goto out;
4614 }
4615
4616 /* In case the driver does not call amend_status_cb() */
4617 qemu_progress_print(0.f, 0);
4618 ret = bdrv_amend_options(bs, opts, &amend_status_cb, NULL, force, &err);
4619 qemu_progress_print(100.f, 0);
4620 bdrv_graph_rdunlock_main_loop();
4621
4622 if (ret < 0) {
4623 error_report_err(err);
4624 goto out;
4625 }
4626
4627 out:
4628 qemu_progress_end();
4629
4630 out_no_progress:
4631 blk_unref(blk);
4632 qemu_opts_del(opts);
4633 qemu_opts_free(amend_opts);
4634 g_free(options);
4635
4636 if (ret) {
4637 return 1;
4638 }
4639 return 0;
4640 }
4641
4642 typedef struct BenchData {
4643 BlockBackend *blk;
4644 uint64_t image_size;
4645 bool write;
4646 int bufsize;
4647 int step;
4648 int nrreq;
4649 int n;
4650 int flush_interval;
4651 bool drain_on_flush;
4652 uint8_t *buf;
4653 QEMUIOVector *qiov;
4654
4655 int in_flight;
4656 bool in_flush;
4657 uint64_t offset;
4658 } BenchData;
4659
4660 static void bench_undrained_flush_cb(void *opaque, int ret)
4661 {
4662 if (ret < 0) {
4663 error_report("Failed flush request: %s", strerror(-ret));
4664 exit(EXIT_FAILURE);
4665 }
4666 }
4667
4668 static void bench_cb(void *opaque, int ret)
4669 {
4670 BenchData *b = opaque;
4671 BlockAIOCB *acb;
4672
4673 if (ret < 0) {
4674 error_report("Failed request: %s", strerror(-ret));
4675 exit(EXIT_FAILURE);
4676 }
4677
4678 if (b->in_flush) {
4679 /* Just finished a flush with drained queue: Start next requests */
4680 assert(b->in_flight == 0);
4681 b->in_flush = false;
4682 } else if (b->in_flight > 0) {
4683 int remaining = b->n - b->in_flight;
4684
4685 b->n--;
4686 b->in_flight--;
4687
4688 /* Time for flush? Drain queue if requested, then flush */
4689 if (b->flush_interval && remaining % b->flush_interval == 0) {
4690 if (!b->in_flight || !b->drain_on_flush) {
4691 BlockCompletionFunc *cb;
4692
4693 if (b->drain_on_flush) {
4694 b->in_flush = true;
4695 cb = bench_cb;
4696 } else {
4697 cb = bench_undrained_flush_cb;
4698 }
4699
4700 acb = blk_aio_flush(b->blk, cb, b);
4701 if (!acb) {
4702 error_report("Failed to issue flush request");
4703 exit(EXIT_FAILURE);
4704 }
4705 }
4706 if (b->drain_on_flush) {
4707 return;
4708 }
4709 }
4710 }
4711
4712 while (b->n > b->in_flight && b->in_flight < b->nrreq) {
4713 int64_t offset = b->offset;
4714 /* blk_aio_* might look for completed I/Os and kick bench_cb
4715 * again, so make sure this operation is counted by in_flight
4716 * and b->offset is ready for the next submission.
4717 */
4718 b->in_flight++;
4719 b->offset += b->step;
4720 if (b->image_size <= b->bufsize) {
4721 b->offset = 0;
4722 } else {
4723 b->offset %= b->image_size - b->bufsize;
4724 }
4725 if (b->write) {
4726 acb = blk_aio_pwritev(b->blk, offset, b->qiov, 0, bench_cb, b);
4727 } else {
4728 acb = blk_aio_preadv(b->blk, offset, b->qiov, 0, bench_cb, b);
4729 }
4730 if (!acb) {
4731 error_report("Failed to issue request");
4732 exit(EXIT_FAILURE);
4733 }
4734 }
4735 }
4736
4737 static int img_bench(const img_cmd_t *ccmd, int argc, char **argv)
4738 {
4739 int c, ret = 0;
4740 const char *fmt = NULL, *filename;
4741 bool quiet = false;
4742 bool image_opts = false;
4743 bool is_write = false;
4744 int count = 75000;
4745 int depth = 64;
4746 int64_t offset = 0;
4747 ssize_t bufsize = 4096;
4748 int pattern = 0;
4749 ssize_t step = 0;
4750 int flush_interval = 0;
4751 bool drain_on_flush = true;
4752 int64_t image_size;
4753 BlockBackend *blk = NULL;
4754 BenchData data = {};
4755 int flags = 0;
4756 bool writethrough = false;
4757 struct timeval t1, t2;
4758 int i;
4759 bool force_share = false;
4760 size_t buf_size = 0;
4761
4762 for (;;) {
4763 static const struct option long_options[] = {
4764 {"help", no_argument, 0, 'h'},
4765 {"format", required_argument, 0, 'f'},
4766 {"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
4767 {"cache", required_argument, 0, 't'},
4768 {"count", required_argument, 0, 'c'},
4769 {"depth", required_argument, 0, 'd'},
4770 {"offset", required_argument, 0, 'o'},
4771 {"buffer-size", required_argument, 0, 's'},
4772 {"step-size", required_argument, 0, 'S'},
4773 {"write", no_argument, 0, 'w'},
4774 {"pattern", required_argument, 0, OPTION_PATTERN},
4775 {"flush-interval", required_argument, 0, OPTION_FLUSH_INTERVAL},
4776 {"no-drain", no_argument, 0, OPTION_NO_DRAIN},
4777 {"aio", required_argument, 0, 'i'},
4778 {"native", no_argument, 0, 'n'},
4779 {"force-share", no_argument, 0, 'U'},
4780 {"quiet", no_argument, 0, 'q'},
4781 {"object", required_argument, 0, OPTION_OBJECT},
4782 {0, 0, 0, 0}
4783 };
4784 c = getopt_long(argc, argv, "hf:t:c:d:o:s:S:wi:nUq",
4785 long_options, NULL);
4786 if (c == -1) {
4787 break;
4788 }
4789
4790 switch (c) {
4791 case 'h':
4792 cmd_help(ccmd, "[-f FMT | --image-opts] [-t CACHE]\n"
4793 " [-c COUNT] [-d DEPTH] [-o OFFSET] [-s BUFFER_SIZE] [-S STEP_SIZE]\n"
4794 " [-w [--pattern PATTERN] [--flush-interval INTERVAL [--no-drain]]]\n"
4795 " [-i AIO] [-n] [-U] [-q] FILE\n"
4796 ,
4797 " -f, --format FMT\n"
4798 " specify FILE format explicitly\n"
4799 " --image-opts\n"
4800 " indicates that FILE is a complete image specification\n"
4801 " instead of a file name (incompatible with --format)\n"
4802 " -t, --cache CACHE\n"
4803 " cache mode for FILE (default: " BDRV_DEFAULT_CACHE ")\n"
4804 " -c, --count COUNT\n"
4805 " number of I/O requests to perform\n"
4806 " -d, --depth DEPTH\n"
4807 " number of requests to perform in parallel\n"
4808 " -o, --offset OFFSET\n"
4809 " start first request at this OFFSET\n"
4810 " -s, --buffer-size BUFFER_SIZE[bkKMGTPE]\n"
4811 " size of each I/O request, with optional multiplier suffix\n"
4812 " (powers of 1024, default is 4K)\n"
4813 " -S, --step-size STEP_SIZE[bkKMGTPE]\n"
4814 " each next request offset increment, with optional multiplier suffix\n"
4815 " (powers of 1024, default is the same as BUFFER_SIZE)\n"
4816 " -w, --write\n"
4817 " perform write test (default is read)\n"
4818 " --pattern PATTERN\n"
4819 " write this pattern byte instead of zero\n"
4820 " --flush-interval FLUSH_INTERVAL\n"
4821 " issue flush after this number of requests\n"
4822 " --no-drain\n"
4823 " do not wait when flushing pending requests\n"
4824 " -i, --aio AIO\n"
4825 " async-io backend (threads, native, io_uring)\n"
4826 " -n, --native\n"
4827 " use native AIO backend if possible\n"
4828 " -U, --force-share\n"
4829 " open images in shared mode for concurrent access\n"
4830 " -q, --quiet\n"
4831 " quiet mode (produce only error messages if any)\n"
4832 " --object OBJDEF\n"
4833 " defines QEMU user-creatable object\n"
4834 " FILE\n"
4835 " name of the image file, or option string (key=value,..)\n"
4836 " with --image-opts, to operate on\n"
4837 );
4838 break;
4839 case 'f':
4840 fmt = optarg;
4841 break;
4842 case OPTION_IMAGE_OPTS:
4843 image_opts = true;
4844 break;
4845 case 't':
4846 ret = bdrv_parse_cache_mode(optarg, &flags, &writethrough);
4847 if (ret < 0) {
4848 error_report("Invalid cache mode");
4849 ret = -1;
4850 goto out;
4851 }
4852 break;
4853 case 'c':
4854 count = cvtnum_full("request count", optarg, false, 1, INT_MAX);
4855 if (count < 0) {
4856 return 1;
4857 }
4858 break;
4859 case 'd':
4860 depth = cvtnum_full("queue depth", optarg, false, 1, INT_MAX);
4861 if (depth < 0) {
4862 return 1;
4863 }
4864 break;
4865 case 'n':
4866 flags |= BDRV_O_NATIVE_AIO;
4867 break;
4868 case 'i':
4869 ret = bdrv_parse_aio(optarg, &flags);
4870 if (ret < 0) {
4871 error_report("Invalid aio option: %s", optarg);
4872 ret = -1;
4873 goto out;
4874 }
4875 break;
4876 case 'o':
4877 offset = cvtnum("offset", optarg, true);
4878 if (offset < 0) {
4879 return 1;
4880 }
4881 break;
4882 case 's':
4883 bufsize = cvtnum_full("buffer size", optarg, true, 1, INT_MAX);
4884 if (bufsize < 0) {
4885 return 1;
4886 }
4887 break;
4888 case 'S':
4889 step = cvtnum_full("step size", optarg, true, 0, INT_MAX);
4890 if (step < 0) {
4891 return 1;
4892 }
4893 break;
4894 case 'w':
4895 flags |= BDRV_O_RDWR;
4896 is_write = true;
4897 break;
4898 case OPTION_PATTERN:
4899 pattern = cvtnum_full("pattern byte", optarg, false, 0, 0xff);
4900 if (pattern < 0) {
4901 return 1;
4902 }
4903 break;
4904 case OPTION_FLUSH_INTERVAL:
4905 flush_interval = cvtnum_full("flush interval", optarg,
4906 false, 0, INT_MAX);
4907 if (flush_interval < 0) {
4908 return 1;
4909 }
4910 break;
4911 case OPTION_NO_DRAIN:
4912 drain_on_flush = false;
4913 break;
4914 case 'U':
4915 force_share = true;
4916 break;
4917 case 'q':
4918 quiet = true;
4919 break;
4920 case OPTION_OBJECT:
4921 user_creatable_process_cmdline(optarg);
4922 break;
4923 default:
4924 tryhelp(argv[0]);
4925 }
4926 }
4927
4928 if (optind != argc - 1) {
4929 error_exit(argv[0], "Expecting one image file name");
4930 }
4931 filename = argv[argc - 1];
4932
4933 if (!is_write && flush_interval) {
4934 error_report("--flush-interval is only available in write tests");
4935 ret = -1;
4936 goto out;
4937 }
4938 if (flush_interval && flush_interval < depth) {
4939 error_report("Flush interval can't be smaller than depth");
4940 ret = -1;
4941 goto out;
4942 }
4943
4944 blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
4945 force_share);
4946 if (!blk) {
4947 ret = -1;
4948 goto out;
4949 }
4950
4951 image_size = blk_getlength(blk);
4952 if (image_size < 0) {
4953 ret = image_size;
4954 goto out;
4955 }
4956
4957 data = (BenchData) {
4958 .blk = blk,
4959 .image_size = image_size,
4960 .bufsize = bufsize,
4961 .step = step ?: bufsize,
4962 .nrreq = depth,
4963 .n = count,
4964 .offset = offset,
4965 .write = is_write,
4966 .flush_interval = flush_interval,
4967 .drain_on_flush = drain_on_flush,
4968 };
4969 printf("Sending %d %s requests, %d bytes each, %d in parallel "
4970 "(starting at offset %" PRId64 ", step size %d)\n",
4971 data.n, data.write ? "write" : "read", data.bufsize, data.nrreq,
4972 data.offset, data.step);
4973 if (flush_interval) {
4974 printf("Sending flush every %d requests\n", flush_interval);
4975 }
4976
4977 buf_size = data.nrreq * data.bufsize;
4978 data.buf = blk_blockalign(blk, buf_size);
4979 memset(data.buf, pattern, data.nrreq * data.bufsize);
4980
4981 blk_register_buf(blk, data.buf, buf_size, &error_fatal);
4982
4983 data.qiov = g_new(QEMUIOVector, data.nrreq);
4984 for (i = 0; i < data.nrreq; i++) {
4985 qemu_iovec_init(&data.qiov[i], 1);
4986 qemu_iovec_add(&data.qiov[i],
4987 data.buf + i * data.bufsize, data.bufsize);
4988 }
4989
4990 gettimeofday(&t1, NULL);
4991 bench_cb(&data, 0);
4992
4993 while (data.n > 0) {
4994 main_loop_wait(false);
4995 }
4996 gettimeofday(&t2, NULL);
4997
4998 printf("Run completed in %3.3f seconds.\n",
4999 (t2.tv_sec - t1.tv_sec)
5000 + ((double)(t2.tv_usec - t1.tv_usec) / 1000000));
Showing first 5,000 of 6,049 lines. View raw