]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/partition/repart.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Simplify random number selection
[thirdparty/systemd.git] / src / partition / repart.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #if HAVE_VALGRIND_MEMCHECK_H
4 #include <valgrind/memcheck.h>
5 #endif
6
7 #include <fcntl.h>
8 #include <getopt.h>
9 #include <linux/fs.h>
10 #include <linux/loop.h>
11 #include <sys/file.h>
12 #include <sys/ioctl.h>
13 #include <sys/stat.h>
14
15 #include "sd-device.h"
16 #include "sd-id128.h"
17
18 #include "alloc-util.h"
19 #include "blkid-util.h"
20 #include "blockdev-util.h"
21 #include "btrfs-util.h"
22 #include "chase-symlinks.h"
23 #include "conf-files.h"
24 #include "conf-parser.h"
25 #include "cryptsetup-util.h"
26 #include "def.h"
27 #include "devnum-util.h"
28 #include "dirent-util.h"
29 #include "efivars.h"
30 #include "errno-util.h"
31 #include "fd-util.h"
32 #include "fdisk-util.h"
33 #include "fileio.h"
34 #include "format-table.h"
35 #include "format-util.h"
36 #include "fs-util.h"
37 #include "glyph-util.h"
38 #include "gpt.h"
39 #include "hexdecoct.h"
40 #include "hmac.h"
41 #include "id128-util.h"
42 #include "json.h"
43 #include "list.h"
44 #include "loop-util.h"
45 #include "main-func.h"
46 #include "mkdir.h"
47 #include "mkfs-util.h"
48 #include "mount-util.h"
49 #include "mountpoint-util.h"
50 #include "parse-argument.h"
51 #include "parse-helpers.h"
52 #include "pretty-print.h"
53 #include "proc-cmdline.h"
54 #include "process-util.h"
55 #include "random-util.h"
56 #include "resize-fs.h"
57 #include "sort-util.h"
58 #include "specifier.h"
59 #include "stdio-util.h"
60 #include "string-table.h"
61 #include "string-util.h"
62 #include "strv.h"
63 #include "sync-util.h"
64 #include "terminal-util.h"
65 #include "tpm2-util.h"
66 #include "user-util.h"
67 #include "utf8.h"
68
69 /* If not configured otherwise use a minimal partition size of 10M */
70 #define DEFAULT_MIN_SIZE (10*1024*1024)
71
72 /* Hard lower limit for new partition sizes */
73 #define HARD_MIN_SIZE 4096
74
75 /* libfdisk takes off slightly more than 1M of the disk size when creating a GPT disk label */
76 #define GPT_METADATA_SIZE (1044*1024)
77
78 /* LUKS2 takes off 16M of the partition size with its metadata by default */
79 #define LUKS2_METADATA_SIZE (16*1024*1024)
80
81 /* Note: When growing and placing new partitions we always align to 4K sector size. It's how newer hard disks
82 * are designed, and if everything is aligned to that performance is best. And for older hard disks with 512B
83 * sector size devices were generally assumed to have an even number of sectors, hence at the worst we'll
84 * waste 3K per partition, which is probably fine. */
85
86 static enum {
87 EMPTY_REFUSE, /* refuse empty disks, never create a partition table */
88 EMPTY_ALLOW, /* allow empty disks, create partition table if necessary */
89 EMPTY_REQUIRE, /* require an empty disk, create a partition table */
90 EMPTY_FORCE, /* make disk empty, erase everything, create a partition table always */
91 EMPTY_CREATE, /* create disk as loopback file, create a partition table always */
92 } arg_empty = EMPTY_REFUSE;
93
94 static bool arg_dry_run = true;
95 static const char *arg_node = NULL;
96 static char *arg_root = NULL;
97 static char *arg_image = NULL;
98 static char *arg_definitions = NULL;
99 static bool arg_discard = true;
100 static bool arg_can_factory_reset = false;
101 static int arg_factory_reset = -1;
102 static sd_id128_t arg_seed = SD_ID128_NULL;
103 static bool arg_randomize = false;
104 static int arg_pretty = -1;
105 static uint64_t arg_size = UINT64_MAX;
106 static bool arg_size_auto = false;
107 static JsonFormatFlags arg_json_format_flags = JSON_FORMAT_OFF;
108 static PagerFlags arg_pager_flags = 0;
109 static bool arg_legend = true;
110 static void *arg_key = NULL;
111 static size_t arg_key_size = 0;
112 static char *arg_tpm2_device = NULL;
113 static uint32_t arg_tpm2_pcr_mask = UINT32_MAX;
114
115 STATIC_DESTRUCTOR_REGISTER(arg_root, freep);
116 STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
117 STATIC_DESTRUCTOR_REGISTER(arg_definitions, freep);
118 STATIC_DESTRUCTOR_REGISTER(arg_key, erase_and_freep);
119 STATIC_DESTRUCTOR_REGISTER(arg_tpm2_device, freep);
120
121 typedef struct Partition Partition;
122 typedef struct FreeArea FreeArea;
123 typedef struct Context Context;
124
125 typedef enum EncryptMode {
126 ENCRYPT_OFF,
127 ENCRYPT_KEY_FILE,
128 ENCRYPT_TPM2,
129 ENCRYPT_KEY_FILE_TPM2,
130 _ENCRYPT_MODE_MAX,
131 _ENCRYPT_MODE_INVALID = -EINVAL,
132 } EncryptMode;
133
134 struct Partition {
135 char *definition_path;
136
137 sd_id128_t type_uuid;
138 sd_id128_t current_uuid, new_uuid;
139 char *current_label, *new_label;
140
141 bool dropped;
142 bool factory_reset;
143 int32_t priority;
144
145 uint32_t weight, padding_weight;
146
147 uint64_t current_size, new_size;
148 uint64_t size_min, size_max;
149
150 uint64_t current_padding, new_padding;
151 uint64_t padding_min, padding_max;
152
153 uint64_t partno;
154 uint64_t offset;
155
156 struct fdisk_partition *current_partition;
157 struct fdisk_partition *new_partition;
158 FreeArea *padding_area;
159 FreeArea *allocated_to_area;
160
161 char *copy_blocks_path;
162 bool copy_blocks_auto;
163 int copy_blocks_fd;
164 uint64_t copy_blocks_size;
165
166 char *format;
167 char **copy_files;
168 char **make_directories;
169 EncryptMode encrypt;
170
171 uint64_t gpt_flags;
172 int no_auto;
173 int read_only;
174 int growfs;
175
176 LIST_FIELDS(Partition, partitions);
177 };
178
179 #define PARTITION_IS_FOREIGN(p) (!(p)->definition_path)
180 #define PARTITION_EXISTS(p) (!!(p)->current_partition)
181
182 struct FreeArea {
183 Partition *after;
184 uint64_t size;
185 uint64_t allocated;
186 };
187
188 struct Context {
189 LIST_HEAD(Partition, partitions);
190 size_t n_partitions;
191
192 FreeArea **free_areas;
193 size_t n_free_areas;
194
195 uint64_t start, end, total;
196
197 struct fdisk_context *fdisk_context;
198 uint64_t sector_size;
199 uint64_t grain_size;
200
201 sd_id128_t seed;
202 };
203
204 static const char *encrypt_mode_table[_ENCRYPT_MODE_MAX] = {
205 [ENCRYPT_OFF] = "off",
206 [ENCRYPT_KEY_FILE] = "key-file",
207 [ENCRYPT_TPM2] = "tpm2",
208 [ENCRYPT_KEY_FILE_TPM2] = "key-file+tpm2",
209 };
210
211 #if HAVE_LIBCRYPTSETUP
212 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(encrypt_mode, EncryptMode, ENCRYPT_KEY_FILE);
213 #else
214 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_FROM_STRING_WITH_BOOLEAN(encrypt_mode, EncryptMode, ENCRYPT_KEY_FILE);
215 #endif
216
217
218 static uint64_t round_down_size(uint64_t v, uint64_t p) {
219 return (v / p) * p;
220 }
221
222 static uint64_t round_up_size(uint64_t v, uint64_t p) {
223
224 v = DIV_ROUND_UP(v, p);
225
226 if (v > UINT64_MAX / p)
227 return UINT64_MAX; /* overflow */
228
229 return v * p;
230 }
231
232 static Partition *partition_new(void) {
233 Partition *p;
234
235 p = new(Partition, 1);
236 if (!p)
237 return NULL;
238
239 *p = (Partition) {
240 .weight = 1000,
241 .padding_weight = 0,
242 .current_size = UINT64_MAX,
243 .new_size = UINT64_MAX,
244 .size_min = UINT64_MAX,
245 .size_max = UINT64_MAX,
246 .current_padding = UINT64_MAX,
247 .new_padding = UINT64_MAX,
248 .padding_min = UINT64_MAX,
249 .padding_max = UINT64_MAX,
250 .partno = UINT64_MAX,
251 .offset = UINT64_MAX,
252 .copy_blocks_fd = -1,
253 .copy_blocks_size = UINT64_MAX,
254 .no_auto = -1,
255 .read_only = -1,
256 .growfs = -1,
257 };
258
259 return p;
260 }
261
262 static Partition* partition_free(Partition *p) {
263 if (!p)
264 return NULL;
265
266 free(p->current_label);
267 free(p->new_label);
268 free(p->definition_path);
269
270 if (p->current_partition)
271 fdisk_unref_partition(p->current_partition);
272 if (p->new_partition)
273 fdisk_unref_partition(p->new_partition);
274
275 free(p->copy_blocks_path);
276 safe_close(p->copy_blocks_fd);
277
278 free(p->format);
279 strv_free(p->copy_files);
280 strv_free(p->make_directories);
281
282 return mfree(p);
283 }
284
285 static Partition* partition_unlink_and_free(Context *context, Partition *p) {
286 if (!p)
287 return NULL;
288
289 LIST_REMOVE(partitions, context->partitions, p);
290
291 assert(context->n_partitions > 0);
292 context->n_partitions--;
293
294 return partition_free(p);
295 }
296
297 DEFINE_TRIVIAL_CLEANUP_FUNC(Partition*, partition_free);
298
299 static Context *context_new(sd_id128_t seed) {
300 Context *context;
301
302 context = new(Context, 1);
303 if (!context)
304 return NULL;
305
306 *context = (Context) {
307 .start = UINT64_MAX,
308 .end = UINT64_MAX,
309 .total = UINT64_MAX,
310 .seed = seed,
311 };
312
313 return context;
314 }
315
316 static void context_free_free_areas(Context *context) {
317 assert(context);
318
319 for (size_t i = 0; i < context->n_free_areas; i++)
320 free(context->free_areas[i]);
321
322 context->free_areas = mfree(context->free_areas);
323 context->n_free_areas = 0;
324 }
325
326 static Context *context_free(Context *context) {
327 if (!context)
328 return NULL;
329
330 while (context->partitions)
331 partition_unlink_and_free(context, context->partitions);
332 assert(context->n_partitions == 0);
333
334 context_free_free_areas(context);
335
336 if (context->fdisk_context)
337 fdisk_unref_context(context->fdisk_context);
338
339 return mfree(context);
340 }
341
342 DEFINE_TRIVIAL_CLEANUP_FUNC(Context*, context_free);
343
344 static int context_add_free_area(
345 Context *context,
346 uint64_t size,
347 Partition *after) {
348
349 FreeArea *a;
350
351 assert(context);
352 assert(!after || !after->padding_area);
353
354 if (!GREEDY_REALLOC(context->free_areas, context->n_free_areas + 1))
355 return -ENOMEM;
356
357 a = new(FreeArea, 1);
358 if (!a)
359 return -ENOMEM;
360
361 *a = (FreeArea) {
362 .size = size,
363 .after = after,
364 };
365
366 context->free_areas[context->n_free_areas++] = a;
367
368 if (after)
369 after->padding_area = a;
370
371 return 0;
372 }
373
374 static bool context_drop_one_priority(Context *context) {
375 int32_t priority = 0;
376 bool exists = false;
377
378 LIST_FOREACH(partitions, p, context->partitions) {
379 if (p->dropped)
380 continue;
381 if (p->priority < priority)
382 continue;
383 if (p->priority == priority) {
384 exists = exists || PARTITION_EXISTS(p);
385 continue;
386 }
387
388 priority = p->priority;
389 exists = PARTITION_EXISTS(p);
390 }
391
392 /* Refuse to drop partitions with 0 or negative priorities or partitions of priorities that have at
393 * least one existing priority */
394 if (priority <= 0 || exists)
395 return false;
396
397 LIST_FOREACH(partitions, p, context->partitions) {
398 if (p->priority < priority)
399 continue;
400
401 if (p->dropped)
402 continue;
403
404 p->dropped = true;
405 log_info("Can't fit partition %s of priority %" PRIi32 ", dropping.", p->definition_path, p->priority);
406 }
407
408 return true;
409 }
410
411 static uint64_t partition_min_size(Context *context, const Partition *p) {
412 uint64_t sz;
413
414 assert(context);
415 assert(p);
416
417 /* Calculate the disk space we really need at minimum for this partition. If the partition already
418 * exists the current size is what we really need. If it doesn't exist yet refuse to allocate less
419 * than 4K.
420 *
421 * DEFAULT_MIN_SIZE is the default SizeMin= we configure if nothing else is specified. */
422
423 if (PARTITION_IS_FOREIGN(p)) {
424 /* Don't allow changing size of partitions not managed by us */
425 assert(p->current_size != UINT64_MAX);
426 return p->current_size;
427 }
428
429 sz = p->current_size != UINT64_MAX ? p->current_size : HARD_MIN_SIZE;
430
431 if (!PARTITION_EXISTS(p)) {
432 uint64_t d = 0;
433
434 if (p->encrypt != ENCRYPT_OFF)
435 d += round_up_size(LUKS2_METADATA_SIZE, context->grain_size);
436
437 if (p->copy_blocks_size != UINT64_MAX)
438 d += round_up_size(p->copy_blocks_size, context->grain_size);
439 else if (p->format || p->encrypt != ENCRYPT_OFF) {
440 uint64_t f;
441
442 /* If we shall synthesize a file system, take minimal fs size into account (assumed to be 4K if not known) */
443 f = p->format ? round_up_size(minimal_size_by_fs_name(p->format), context->grain_size) : UINT64_MAX;
444 d += f == UINT64_MAX ? context->grain_size : f;
445 }
446
447 if (d > sz)
448 sz = d;
449 }
450
451 return MAX(round_up_size(p->size_min != UINT64_MAX ? p->size_min : DEFAULT_MIN_SIZE, context->grain_size), sz);
452 }
453
454 static uint64_t partition_max_size(const Context *context, const Partition *p) {
455 uint64_t sm;
456
457 /* Calculate how large the partition may become at max. This is generally the configured maximum
458 * size, except when it already exists and is larger than that. In that case it's the existing size,
459 * since we never want to shrink partitions. */
460
461 assert(context);
462 assert(p);
463
464 if (PARTITION_IS_FOREIGN(p)) {
465 /* Don't allow changing size of partitions not managed by us */
466 assert(p->current_size != UINT64_MAX);
467 return p->current_size;
468 }
469
470 sm = round_down_size(p->size_max, context->grain_size);
471
472 if (p->current_size != UINT64_MAX)
473 return MAX(p->current_size, sm);
474
475 return sm;
476 }
477
478 static uint64_t partition_min_size_with_padding(Context *context, const Partition *p) {
479 uint64_t sz;
480
481 /* Calculate the disk space we need for this partition plus any free space coming after it. This
482 * takes user configured padding into account as well as any additional whitespace needed to align
483 * the next partition to 4K again. */
484
485 assert(context);
486 assert(p);
487
488 sz = partition_min_size(context, p);
489
490 if (p->padding_min != UINT64_MAX)
491 sz += p->padding_min;
492
493 if (PARTITION_EXISTS(p)) {
494 /* If the partition wasn't aligned, add extra space so that any we might add will be aligned */
495 assert(p->offset != UINT64_MAX);
496 return round_up_size(p->offset + sz, context->grain_size) - p->offset;
497 }
498
499 /* If this is a new partition we'll place it aligned, hence we just need to round up the required size here */
500 return round_up_size(sz, context->grain_size);
501 }
502
503 static uint64_t free_area_available(const FreeArea *a) {
504 assert(a);
505
506 /* Determines how much of this free area is not allocated yet */
507
508 assert(a->size >= a->allocated);
509 return a->size - a->allocated;
510 }
511
512 static uint64_t free_area_available_for_new_partitions(Context *context, const FreeArea *a) {
513 uint64_t avail;
514
515 assert(context);
516 assert(a);
517
518 /* Similar to free_area_available(), but takes into account that the required size and padding of the
519 * preceding partition is honoured. */
520
521 avail = free_area_available(a);
522 if (a->after) {
523 uint64_t need, space_end, new_end;
524
525 need = partition_min_size_with_padding(context, a->after);
526
527 assert(a->after->offset != UINT64_MAX);
528 assert(a->after->current_size != UINT64_MAX);
529
530 /* Calculate where the free area ends, based on the offset of the partition preceding it */
531 space_end = round_up_size(a->after->offset + a->after->current_size, context->grain_size) + avail;
532
533 /* Calculate where the partition would end when we give it as much as it needs */
534 new_end = round_up_size(a->after->offset + need, context->grain_size);
535
536 /* Calculate saturated difference of the two: that's how much we have free for other partitions */
537 return LESS_BY(space_end, new_end);
538 }
539
540 return avail;
541 }
542
543 static int free_area_compare(FreeArea *const *a, FreeArea *const*b, Context *context) {
544 assert(context);
545
546 return CMP(free_area_available_for_new_partitions(context, *a),
547 free_area_available_for_new_partitions(context, *b));
548 }
549
550 static uint64_t charge_size(Context *context, uint64_t total, uint64_t amount) {
551 assert(context);
552 /* Subtract the specified amount from total, rounding up to multiple of 4K if there's room */
553 assert(amount <= total);
554 return LESS_BY(total, round_up_size(amount, context->grain_size));
555 }
556
557 static uint64_t charge_weight(uint64_t total, uint64_t amount) {
558 assert(amount <= total);
559 return total - amount;
560 }
561
562 static bool context_allocate_partitions(Context *context, uint64_t *ret_largest_free_area) {
563 assert(context);
564
565 /* Sort free areas by size, putting smallest first */
566 typesafe_qsort_r(context->free_areas, context->n_free_areas, free_area_compare, context);
567
568 /* In any case return size of the largest free area (i.e. not the size of all free areas
569 * combined!) */
570 if (ret_largest_free_area)
571 *ret_largest_free_area =
572 context->n_free_areas == 0 ? 0 :
573 free_area_available_for_new_partitions(context, context->free_areas[context->n_free_areas-1]);
574
575 /* A simple first-fit algorithm. We return true if we can fit the partitions in, otherwise false. */
576 LIST_FOREACH(partitions, p, context->partitions) {
577 bool fits = false;
578 uint64_t required;
579 FreeArea *a = NULL;
580
581 /* Skip partitions we already dropped or that already exist */
582 if (p->dropped || PARTITION_EXISTS(p))
583 continue;
584
585 /* How much do we need to fit? */
586 required = partition_min_size_with_padding(context, p);
587 assert(required % context->grain_size == 0);
588
589 for (size_t i = 0; i < context->n_free_areas; i++) {
590 a = context->free_areas[i];
591
592 if (free_area_available_for_new_partitions(context, a) >= required) {
593 fits = true;
594 break;
595 }
596 }
597
598 if (!fits)
599 return false; /* 😢 Oh no! We can't fit this partition into any free area! */
600
601 /* Assign the partition to this free area */
602 p->allocated_to_area = a;
603
604 /* Budget the minimal partition size */
605 a->allocated += required;
606 }
607
608 return true;
609 }
610
611 static int context_sum_weights(Context *context, FreeArea *a, uint64_t *ret) {
612 uint64_t weight_sum = 0;
613
614 assert(context);
615 assert(a);
616 assert(ret);
617
618 /* Determine the sum of the weights of all partitions placed in or before the specified free area */
619
620 LIST_FOREACH(partitions, p, context->partitions) {
621 if (p->padding_area != a && p->allocated_to_area != a)
622 continue;
623
624 if (p->weight > UINT64_MAX - weight_sum)
625 goto overflow_sum;
626 weight_sum += p->weight;
627
628 if (p->padding_weight > UINT64_MAX - weight_sum)
629 goto overflow_sum;
630 weight_sum += p->padding_weight;
631 }
632
633 *ret = weight_sum;
634 return 0;
635
636 overflow_sum:
637 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Combined weight of partition exceeds unsigned 64bit range, refusing.");
638 }
639
640 static int scale_by_weight(uint64_t value, uint64_t weight, uint64_t weight_sum, uint64_t *ret) {
641 assert(weight_sum >= weight);
642 assert(ret);
643
644 if (weight == 0) {
645 *ret = 0;
646 return 0;
647 }
648
649 if (value > UINT64_MAX / weight)
650 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Scaling by weight of partition exceeds unsigned 64bit range, refusing.");
651
652 *ret = value * weight / weight_sum;
653 return 0;
654 }
655
656 typedef enum GrowPartitionPhase {
657 /* The first phase: we charge partitions which need more (according to constraints) than their weight-based share. */
658 PHASE_OVERCHARGE,
659
660 /* The second phase: we charge partitions which need less (according to constraints) than their weight-based share. */
661 PHASE_UNDERCHARGE,
662
663 /* The third phase: we distribute what remains among the remaining partitions, according to the weights */
664 PHASE_DISTRIBUTE,
665
666 _GROW_PARTITION_PHASE_MAX,
667 } GrowPartitionPhase;
668
669 static int context_grow_partitions_phase(
670 Context *context,
671 FreeArea *a,
672 GrowPartitionPhase phase,
673 uint64_t *span,
674 uint64_t *weight_sum) {
675
676 int r;
677
678 assert(context);
679 assert(a);
680
681 /* Now let's look at the intended weights and adjust them taking the minimum space assignments into
682 * account. i.e. if a partition has a small weight but a high minimum space value set it should not
683 * get any additional room from the left-overs. Similar, if two partitions have the same weight they
684 * should get the same space if possible, even if one has a smaller minimum size than the other. */
685 LIST_FOREACH(partitions, p, context->partitions) {
686
687 /* Look only at partitions associated with this free area, i.e. immediately
688 * preceding it, or allocated into it */
689 if (p->allocated_to_area != a && p->padding_area != a)
690 continue;
691
692 if (p->new_size == UINT64_MAX) {
693 bool charge = false, try_again = false;
694 uint64_t share, rsz, xsz;
695
696 /* Calculate how much this space this partition needs if everyone would get
697 * the weight based share */
698 r = scale_by_weight(*span, p->weight, *weight_sum, &share);
699 if (r < 0)
700 return r;
701
702 rsz = partition_min_size(context, p);
703 xsz = partition_max_size(context, p);
704
705 if (phase == PHASE_OVERCHARGE && rsz > share) {
706 /* This partition needs more than its calculated share. Let's assign
707 * it that, and take this partition out of all calculations and start
708 * again. */
709
710 p->new_size = rsz;
711 charge = try_again = true;
712
713 } else if (phase == PHASE_UNDERCHARGE && xsz != UINT64_MAX && xsz < share) {
714 /* This partition accepts less than its calculated
715 * share. Let's assign it that, and take this partition out
716 * of all calculations and start again. */
717
718 p->new_size = xsz;
719 charge = try_again = true;
720
721 } else if (phase == PHASE_DISTRIBUTE) {
722 /* This partition can accept its calculated share. Let's
723 * assign it. There's no need to restart things here since
724 * assigning this shouldn't impact the shares of the other
725 * partitions. */
726
727 if (PARTITION_IS_FOREIGN(p))
728 /* Never change of foreign partitions (i.e. those we don't manage) */
729 p->new_size = p->current_size;
730 else
731 p->new_size = MAX(round_down_size(share, context->grain_size), rsz);
732
733 charge = true;
734 }
735
736 if (charge) {
737 *span = charge_size(context, *span, p->new_size);
738 *weight_sum = charge_weight(*weight_sum, p->weight);
739 }
740
741 if (try_again)
742 return 0; /* try again */
743 }
744
745 if (p->new_padding == UINT64_MAX) {
746 bool charge = false, try_again = false;
747 uint64_t share;
748
749 r = scale_by_weight(*span, p->padding_weight, *weight_sum, &share);
750 if (r < 0)
751 return r;
752
753 if (phase == PHASE_OVERCHARGE && p->padding_min != UINT64_MAX && p->padding_min > share) {
754 p->new_padding = p->padding_min;
755 charge = try_again = true;
756 } else if (phase == PHASE_UNDERCHARGE && p->padding_max != UINT64_MAX && p->padding_max < share) {
757 p->new_padding = p->padding_max;
758 charge = try_again = true;
759 } else if (phase == PHASE_DISTRIBUTE) {
760
761 p->new_padding = round_down_size(share, context->grain_size);
762 if (p->padding_min != UINT64_MAX && p->new_padding < p->padding_min)
763 p->new_padding = p->padding_min;
764
765 charge = true;
766 }
767
768 if (charge) {
769 *span = charge_size(context, *span, p->new_padding);
770 *weight_sum = charge_weight(*weight_sum, p->padding_weight);
771 }
772
773 if (try_again)
774 return 0; /* try again */
775 }
776 }
777
778 return 1; /* done */
779 }
780
781 static int context_grow_partitions_on_free_area(Context *context, FreeArea *a) {
782 uint64_t weight_sum = 0, span;
783 int r;
784
785 assert(context);
786 assert(a);
787
788 r = context_sum_weights(context, a, &weight_sum);
789 if (r < 0)
790 return r;
791
792 /* Let's calculate the total area covered by this free area and the partition before it */
793 span = a->size;
794 if (a->after) {
795 assert(a->after->offset != UINT64_MAX);
796 assert(a->after->current_size != UINT64_MAX);
797
798 span += round_up_size(a->after->offset + a->after->current_size, context->grain_size) - a->after->offset;
799 }
800
801 for (GrowPartitionPhase phase = 0; phase < _GROW_PARTITION_PHASE_MAX;) {
802 r = context_grow_partitions_phase(context, a, phase, &span, &weight_sum);
803 if (r < 0)
804 return r;
805 if (r == 0) /* not done yet, re-run this phase */
806 continue;
807
808 phase++; /* got to next phase */
809 }
810
811 /* We still have space left over? Donate to preceding partition if we have one */
812 if (span > 0 && a->after && !PARTITION_IS_FOREIGN(a->after)) {
813 uint64_t m, xsz;
814
815 assert(a->after->new_size != UINT64_MAX);
816
817 /* Calculate new size and align (but ensure this doesn't shrink the size) */
818 m = MAX(a->after->new_size, round_down_size(a->after->new_size + span, context->grain_size));
819
820 xsz = partition_max_size(context, a->after);
821 if (xsz != UINT64_MAX && m > xsz)
822 m = xsz;
823
824 span = charge_size(context, span, m - a->after->new_size);
825 a->after->new_size = m;
826 }
827
828 /* What? Even still some space left (maybe because there was no preceding partition, or it had a
829 * size limit), then let's donate it to whoever wants it. */
830 if (span > 0)
831 LIST_FOREACH(partitions, p, context->partitions) {
832 uint64_t m, xsz;
833
834 if (p->allocated_to_area != a)
835 continue;
836
837 if (PARTITION_IS_FOREIGN(p))
838 continue;
839
840 assert(p->new_size != UINT64_MAX);
841 m = MAX(p->new_size, round_down_size(p->new_size + span, context->grain_size));
842
843 xsz = partition_max_size(context, p);
844 if (xsz != UINT64_MAX && m > xsz)
845 m = xsz;
846
847 span = charge_size(context, span, m - p->new_size);
848 p->new_size = m;
849
850 if (span == 0)
851 break;
852 }
853
854 /* Yuck, still no one? Then make it padding */
855 if (span > 0 && a->after) {
856 assert(a->after->new_padding != UINT64_MAX);
857 a->after->new_padding += span;
858 }
859
860 return 0;
861 }
862
863 static int context_grow_partitions(Context *context) {
864 int r;
865
866 assert(context);
867
868 for (size_t i = 0; i < context->n_free_areas; i++) {
869 r = context_grow_partitions_on_free_area(context, context->free_areas[i]);
870 if (r < 0)
871 return r;
872 }
873
874 /* All existing partitions that have no free space after them can't change size */
875 LIST_FOREACH(partitions, p, context->partitions) {
876 if (p->dropped)
877 continue;
878
879 if (!PARTITION_EXISTS(p) || p->padding_area) {
880 /* The algorithm above must have initialized this already */
881 assert(p->new_size != UINT64_MAX);
882 continue;
883 }
884
885 assert(p->new_size == UINT64_MAX);
886 p->new_size = p->current_size;
887
888 assert(p->new_padding == UINT64_MAX);
889 p->new_padding = p->current_padding;
890 }
891
892 return 0;
893 }
894
895 static void context_place_partitions(Context *context) {
896 uint64_t partno = 0;
897
898 assert(context);
899
900 /* Determine next partition number to assign */
901 LIST_FOREACH(partitions, p, context->partitions) {
902 if (!PARTITION_EXISTS(p))
903 continue;
904
905 assert(p->partno != UINT64_MAX);
906 if (p->partno >= partno)
907 partno = p->partno + 1;
908 }
909
910 for (size_t i = 0; i < context->n_free_areas; i++) {
911 FreeArea *a = context->free_areas[i];
912 _unused_ uint64_t left;
913 uint64_t start;
914
915 if (a->after) {
916 assert(a->after->offset != UINT64_MAX);
917 assert(a->after->new_size != UINT64_MAX);
918 assert(a->after->new_padding != UINT64_MAX);
919
920 start = a->after->offset + a->after->new_size + a->after->new_padding;
921 } else
922 start = context->start;
923
924 start = round_up_size(start, context->grain_size);
925 left = a->size;
926
927 LIST_FOREACH(partitions, p, context->partitions) {
928 if (p->allocated_to_area != a)
929 continue;
930
931 p->offset = start;
932 p->partno = partno++;
933
934 assert(left >= p->new_size);
935 start += p->new_size;
936 left -= p->new_size;
937
938 assert(left >= p->new_padding);
939 start += p->new_padding;
940 left -= p->new_padding;
941 }
942 }
943 }
944
945 static int config_parse_type(
946 const char *unit,
947 const char *filename,
948 unsigned line,
949 const char *section,
950 unsigned section_line,
951 const char *lvalue,
952 int ltype,
953 const char *rvalue,
954 void *data,
955 void *userdata) {
956
957 sd_id128_t *type_uuid = data;
958 int r;
959
960 assert(rvalue);
961 assert(type_uuid);
962
963 r = gpt_partition_type_uuid_from_string(rvalue, type_uuid);
964 if (r < 0)
965 return log_syntax(unit, LOG_ERR, filename, line, r, "Failed to parse partition type: %s", rvalue);
966
967 return 0;
968 }
969
970 static int config_parse_label(
971 const char *unit,
972 const char *filename,
973 unsigned line,
974 const char *section,
975 unsigned section_line,
976 const char *lvalue,
977 int ltype,
978 const char *rvalue,
979 void *data,
980 void *userdata) {
981
982 _cleanup_free_ char *resolved = NULL;
983 char **label = data;
984 int r;
985
986 assert(rvalue);
987 assert(label);
988
989 /* Nota bene: the empty label is a totally valid one. Let's hence not follow our usual rule of
990 * assigning the empty string to reset to default here, but really accept it as label to set. */
991
992 r = specifier_printf(rvalue, GPT_LABEL_MAX, system_and_tmp_specifier_table, arg_root, NULL, &resolved);
993 if (r < 0) {
994 log_syntax(unit, LOG_WARNING, filename, line, r,
995 "Failed to expand specifiers in Label=, ignoring: %s", rvalue);
996 return 0;
997 }
998
999 if (!utf8_is_valid(resolved)) {
1000 log_syntax(unit, LOG_WARNING, filename, line, 0,
1001 "Partition label not valid UTF-8, ignoring: %s", rvalue);
1002 return 0;
1003 }
1004
1005 r = gpt_partition_label_valid(resolved);
1006 if (r < 0) {
1007 log_syntax(unit, LOG_WARNING, filename, line, r,
1008 "Failed to check if string is valid as GPT partition label, ignoring: \"%s\" (from \"%s\")",
1009 resolved, rvalue);
1010 return 0;
1011 }
1012 if (!r) {
1013 log_syntax(unit, LOG_WARNING, filename, line, 0,
1014 "Partition label too long for GPT table, ignoring: \"%s\" (from \"%s\")",
1015 resolved, rvalue);
1016 return 0;
1017 }
1018
1019 free_and_replace(*label, resolved);
1020 return 0;
1021 }
1022
1023 static int config_parse_weight(
1024 const char *unit,
1025 const char *filename,
1026 unsigned line,
1027 const char *section,
1028 unsigned section_line,
1029 const char *lvalue,
1030 int ltype,
1031 const char *rvalue,
1032 void *data,
1033 void *userdata) {
1034
1035 uint32_t *priority = data, v;
1036 int r;
1037
1038 assert(rvalue);
1039 assert(priority);
1040
1041 r = safe_atou32(rvalue, &v);
1042 if (r < 0) {
1043 log_syntax(unit, LOG_WARNING, filename, line, r,
1044 "Failed to parse weight value, ignoring: %s", rvalue);
1045 return 0;
1046 }
1047
1048 if (v > 1000U*1000U) {
1049 log_syntax(unit, LOG_WARNING, filename, line, 0,
1050 "Weight needs to be in range 0…10000000, ignoring: %" PRIu32, v);
1051 return 0;
1052 }
1053
1054 *priority = v;
1055 return 0;
1056 }
1057
1058 static int config_parse_size4096(
1059 const char *unit,
1060 const char *filename,
1061 unsigned line,
1062 const char *section,
1063 unsigned section_line,
1064 const char *lvalue,
1065 int ltype,
1066 const char *rvalue,
1067 void *data,
1068 void *userdata) {
1069
1070 uint64_t *sz = data, parsed;
1071 int r;
1072
1073 assert(rvalue);
1074 assert(data);
1075
1076 r = parse_size(rvalue, 1024, &parsed);
1077 if (r < 0)
1078 return log_syntax(unit, LOG_ERR, filename, line, r,
1079 "Failed to parse size value: %s", rvalue);
1080
1081 if (ltype > 0)
1082 *sz = round_up_size(parsed, 4096);
1083 else if (ltype < 0)
1084 *sz = round_down_size(parsed, 4096);
1085 else
1086 *sz = parsed;
1087
1088 if (*sz != parsed)
1089 log_syntax(unit, LOG_NOTICE, filename, line, r, "Rounded %s= size %" PRIu64 " → %" PRIu64 ", a multiple of 4096.", lvalue, parsed, *sz);
1090
1091 return 0;
1092 }
1093
1094 static int config_parse_fstype(
1095 const char *unit,
1096 const char *filename,
1097 unsigned line,
1098 const char *section,
1099 unsigned section_line,
1100 const char *lvalue,
1101 int ltype,
1102 const char *rvalue,
1103 void *data,
1104 void *userdata) {
1105
1106 char **fstype = data;
1107
1108 assert(rvalue);
1109 assert(data);
1110
1111 if (!filename_is_valid(rvalue))
1112 return log_syntax(unit, LOG_ERR, filename, line, 0,
1113 "File system type is not valid, refusing: %s", rvalue);
1114
1115 return free_and_strdup_warn(fstype, rvalue);
1116 }
1117
1118 static int config_parse_copy_files(
1119 const char *unit,
1120 const char *filename,
1121 unsigned line,
1122 const char *section,
1123 unsigned section_line,
1124 const char *lvalue,
1125 int ltype,
1126 const char *rvalue,
1127 void *data,
1128 void *userdata) {
1129
1130 _cleanup_free_ char *source = NULL, *buffer = NULL, *resolved_source = NULL, *resolved_target = NULL;
1131 const char *p = rvalue, *target;
1132 Partition *partition = data;
1133 int r;
1134
1135 assert(rvalue);
1136 assert(partition);
1137
1138 r = extract_first_word(&p, &source, ":", EXTRACT_CUNESCAPE|EXTRACT_DONT_COALESCE_SEPARATORS);
1139 if (r < 0)
1140 return log_syntax(unit, LOG_ERR, filename, line, r, "Failed to extract source path: %s", rvalue);
1141 if (r == 0) {
1142 log_syntax(unit, LOG_WARNING, filename, line, 0, "No argument specified: %s", rvalue);
1143 return 0;
1144 }
1145
1146 r = extract_first_word(&p, &buffer, ":", EXTRACT_CUNESCAPE|EXTRACT_DONT_COALESCE_SEPARATORS);
1147 if (r < 0)
1148 return log_syntax(unit, LOG_ERR, filename, line, r, "Failed to extract target path: %s", rvalue);
1149 if (r == 0)
1150 target = source; /* No target, then it's the same as the source */
1151 else
1152 target = buffer;
1153
1154 if (!isempty(p))
1155 return log_syntax(unit, LOG_ERR, filename, line, SYNTHETIC_ERRNO(EINVAL), "Too many arguments: %s", rvalue);
1156
1157 r = specifier_printf(source, PATH_MAX-1, system_and_tmp_specifier_table, arg_root, NULL, &resolved_source);
1158 if (r < 0) {
1159 log_syntax(unit, LOG_WARNING, filename, line, r,
1160 "Failed to expand specifiers in CopyFiles= source, ignoring: %s", rvalue);
1161 return 0;
1162 }
1163
1164 r = path_simplify_and_warn(resolved_source, PATH_CHECK_ABSOLUTE, unit, filename, line, lvalue);
1165 if (r < 0)
1166 return 0;
1167
1168 r = specifier_printf(target, PATH_MAX-1, system_and_tmp_specifier_table, arg_root, NULL, &resolved_target);
1169 if (r < 0) {
1170 log_syntax(unit, LOG_WARNING, filename, line, r,
1171 "Failed to expand specifiers in CopyFiles= target, ignoring: %s", resolved_target);
1172 return 0;
1173 }
1174
1175 r = path_simplify_and_warn(resolved_target, PATH_CHECK_ABSOLUTE, unit, filename, line, lvalue);
1176 if (r < 0)
1177 return 0;
1178
1179 r = strv_consume_pair(&partition->copy_files, TAKE_PTR(resolved_source), TAKE_PTR(resolved_target));
1180 if (r < 0)
1181 return log_oom();
1182
1183 return 0;
1184 }
1185
1186 static int config_parse_copy_blocks(
1187 const char *unit,
1188 const char *filename,
1189 unsigned line,
1190 const char *section,
1191 unsigned section_line,
1192 const char *lvalue,
1193 int ltype,
1194 const char *rvalue,
1195 void *data,
1196 void *userdata) {
1197
1198 _cleanup_free_ char *d = NULL;
1199 Partition *partition = data;
1200 int r;
1201
1202 assert(rvalue);
1203 assert(partition);
1204
1205 if (isempty(rvalue)) {
1206 partition->copy_blocks_path = mfree(partition->copy_blocks_path);
1207 partition->copy_blocks_auto = false;
1208 return 0;
1209 }
1210
1211 if (streq(rvalue, "auto")) {
1212 partition->copy_blocks_path = mfree(partition->copy_blocks_path);
1213 partition->copy_blocks_auto = true;
1214 return 0;
1215 }
1216
1217 r = specifier_printf(rvalue, PATH_MAX-1, system_and_tmp_specifier_table, arg_root, NULL, &d);
1218 if (r < 0) {
1219 log_syntax(unit, LOG_WARNING, filename, line, r,
1220 "Failed to expand specifiers in CopyBlocks= source path, ignoring: %s", rvalue);
1221 return 0;
1222 }
1223
1224 r = path_simplify_and_warn(d, PATH_CHECK_ABSOLUTE, unit, filename, line, lvalue);
1225 if (r < 0)
1226 return 0;
1227
1228 free_and_replace(partition->copy_blocks_path, d);
1229 partition->copy_blocks_auto = false;
1230 return 0;
1231 }
1232
1233 static int config_parse_make_dirs(
1234 const char *unit,
1235 const char *filename,
1236 unsigned line,
1237 const char *section,
1238 unsigned section_line,
1239 const char *lvalue,
1240 int ltype,
1241 const char *rvalue,
1242 void *data,
1243 void *userdata) {
1244
1245 Partition *partition = data;
1246 const char *p = rvalue;
1247 int r;
1248
1249 assert(rvalue);
1250 assert(partition);
1251
1252 for (;;) {
1253 _cleanup_free_ char *word = NULL, *d = NULL;
1254
1255 r = extract_first_word(&p, &word, NULL, EXTRACT_UNQUOTE);
1256 if (r == -ENOMEM)
1257 return log_oom();
1258 if (r < 0) {
1259 log_syntax(unit, LOG_WARNING, filename, line, r, "Invalid syntax, ignoring: %s", rvalue);
1260 return 0;
1261 }
1262 if (r == 0)
1263 return 0;
1264
1265 r = specifier_printf(word, PATH_MAX-1, system_and_tmp_specifier_table, arg_root, NULL, &d);
1266 if (r < 0) {
1267 log_syntax(unit, LOG_WARNING, filename, line, r,
1268 "Failed to expand specifiers in MakeDirectories= parameter, ignoring: %s", word);
1269 continue;
1270 }
1271
1272 r = path_simplify_and_warn(d, PATH_CHECK_ABSOLUTE, unit, filename, line, lvalue);
1273 if (r < 0)
1274 continue;
1275
1276 r = strv_consume(&partition->make_directories, TAKE_PTR(d));
1277 if (r < 0)
1278 return log_oom();
1279 }
1280 }
1281
1282 static DEFINE_CONFIG_PARSE_ENUM_WITH_DEFAULT(config_parse_encrypt, encrypt_mode, EncryptMode, ENCRYPT_OFF, "Invalid encryption mode");
1283
1284 static int config_parse_gpt_flags(
1285 const char *unit,
1286 const char *filename,
1287 unsigned line,
1288 const char *section,
1289 unsigned section_line,
1290 const char *lvalue,
1291 int ltype,
1292 const char *rvalue,
1293 void *data,
1294 void *userdata) {
1295
1296 uint64_t *gpt_flags = data;
1297 int r;
1298
1299 assert(rvalue);
1300 assert(gpt_flags);
1301
1302 r = safe_atou64(rvalue, gpt_flags);
1303 if (r < 0) {
1304 log_syntax(unit, LOG_WARNING, filename, line, r,
1305 "Failed to parse Flags= value, ignoring: %s", rvalue);
1306 return 0;
1307 }
1308
1309 return 0;
1310 }
1311
1312 static int partition_read_definition(Partition *p, const char *path) {
1313
1314 ConfigTableItem table[] = {
1315 { "Partition", "Type", config_parse_type, 0, &p->type_uuid },
1316 { "Partition", "Label", config_parse_label, 0, &p->new_label },
1317 { "Partition", "UUID", config_parse_id128, 0, &p->new_uuid },
1318 { "Partition", "Priority", config_parse_int32, 0, &p->priority },
1319 { "Partition", "Weight", config_parse_weight, 0, &p->weight },
1320 { "Partition", "PaddingWeight", config_parse_weight, 0, &p->padding_weight },
1321 { "Partition", "SizeMinBytes", config_parse_size4096, 1, &p->size_min },
1322 { "Partition", "SizeMaxBytes", config_parse_size4096, -1, &p->size_max },
1323 { "Partition", "PaddingMinBytes", config_parse_size4096, 1, &p->padding_min },
1324 { "Partition", "PaddingMaxBytes", config_parse_size4096, -1, &p->padding_max },
1325 { "Partition", "FactoryReset", config_parse_bool, 0, &p->factory_reset },
1326 { "Partition", "CopyBlocks", config_parse_copy_blocks, 0, p },
1327 { "Partition", "Format", config_parse_fstype, 0, &p->format },
1328 { "Partition", "CopyFiles", config_parse_copy_files, 0, p },
1329 { "Partition", "MakeDirectories", config_parse_make_dirs, 0, p },
1330 { "Partition", "Encrypt", config_parse_encrypt, 0, &p->encrypt },
1331 { "Partition", "Flags", config_parse_gpt_flags, 0, &p->gpt_flags },
1332 { "Partition", "ReadOnly", config_parse_tristate, 0, &p->read_only },
1333 { "Partition", "NoAuto", config_parse_tristate, 0, &p->no_auto },
1334 { "Partition", "GrowFileSystem", config_parse_tristate, 0, &p->growfs },
1335 {}
1336 };
1337 int r;
1338
1339 r = config_parse(NULL, path, NULL,
1340 "Partition\0",
1341 config_item_table_lookup, table,
1342 CONFIG_PARSE_WARN,
1343 p,
1344 NULL);
1345 if (r < 0)
1346 return r;
1347
1348 if (p->size_min != UINT64_MAX && p->size_max != UINT64_MAX && p->size_min > p->size_max)
1349 return log_syntax(NULL, LOG_ERR, path, 1, SYNTHETIC_ERRNO(EINVAL),
1350 "SizeMinBytes= larger than SizeMaxBytes=, refusing.");
1351
1352 if (p->padding_min != UINT64_MAX && p->padding_max != UINT64_MAX && p->padding_min > p->padding_max)
1353 return log_syntax(NULL, LOG_ERR, path, 1, SYNTHETIC_ERRNO(EINVAL),
1354 "PaddingMinBytes= larger than PaddingMaxBytes=, refusing.");
1355
1356 if (sd_id128_is_null(p->type_uuid))
1357 return log_syntax(NULL, LOG_ERR, path, 1, SYNTHETIC_ERRNO(EINVAL),
1358 "Type= not defined, refusing.");
1359
1360 if ((p->copy_blocks_path || p->copy_blocks_auto) &&
1361 (p->format || !strv_isempty(p->copy_files) || !strv_isempty(p->make_directories)))
1362 return log_syntax(NULL, LOG_ERR, path, 1, SYNTHETIC_ERRNO(EINVAL),
1363 "Format=/CopyFiles=/MakeDirectories= and CopyBlocks= cannot be combined, refusing.");
1364
1365 if ((!strv_isempty(p->copy_files) || !strv_isempty(p->make_directories)) && streq_ptr(p->format, "swap"))
1366 return log_syntax(NULL, LOG_ERR, path, 1, SYNTHETIC_ERRNO(EINVAL),
1367 "Format=swap and CopyFiles= cannot be combined, refusing.");
1368
1369 if (!p->format && (!strv_isempty(p->copy_files) || !strv_isempty(p->make_directories) || (p->encrypt != ENCRYPT_OFF && !(p->copy_blocks_path || p->copy_blocks_auto)))) {
1370 /* Pick "ext4" as file system if we are configured to copy files or encrypt the device */
1371 p->format = strdup("ext4");
1372 if (!p->format)
1373 return log_oom();
1374 }
1375
1376 /* Verity partitions are read only, let's imply the RO flag hence, unless explicitly configured otherwise. */
1377 if ((gpt_partition_type_is_root_verity(p->type_uuid) ||
1378 gpt_partition_type_is_usr_verity(p->type_uuid)) &&
1379 p->read_only < 0)
1380 p->read_only = true;
1381
1382 /* Default to "growfs" on, unless read-only */
1383 if (gpt_partition_type_knows_growfs(p->type_uuid) &&
1384 p->read_only <= 0)
1385 p->growfs = true;
1386
1387 return 0;
1388 }
1389
1390 static int context_read_definitions(
1391 Context *context,
1392 const char *directory,
1393 const char *root) {
1394
1395 _cleanup_strv_free_ char **files = NULL;
1396 Partition *last = NULL;
1397 int r;
1398
1399 assert(context);
1400
1401 if (directory)
1402 r = conf_files_list_strv(&files, ".conf", NULL, CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED, (const char**) STRV_MAKE(directory));
1403 else
1404 r = conf_files_list_strv(&files, ".conf", root, CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED, (const char**) CONF_PATHS_STRV("repart.d"));
1405 if (r < 0)
1406 return log_error_errno(r, "Failed to enumerate *.conf files: %m");
1407
1408 STRV_FOREACH(f, files) {
1409 _cleanup_(partition_freep) Partition *p = NULL;
1410
1411 p = partition_new();
1412 if (!p)
1413 return log_oom();
1414
1415 p->definition_path = strdup(*f);
1416 if (!p->definition_path)
1417 return log_oom();
1418
1419 r = partition_read_definition(p, *f);
1420 if (r < 0)
1421 return r;
1422
1423 LIST_INSERT_AFTER(partitions, context->partitions, last, p);
1424 last = TAKE_PTR(p);
1425 context->n_partitions++;
1426 }
1427
1428 return 0;
1429 }
1430
1431 static int determine_current_padding(
1432 struct fdisk_context *c,
1433 struct fdisk_table *t,
1434 struct fdisk_partition *p,
1435 uint64_t secsz,
1436 uint64_t grainsz,
1437 uint64_t *ret) {
1438
1439 size_t n_partitions;
1440 uint64_t offset, next = UINT64_MAX;
1441
1442 assert(c);
1443 assert(t);
1444 assert(p);
1445
1446 if (!fdisk_partition_has_end(p))
1447 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Partition has no end!");
1448
1449 offset = fdisk_partition_get_end(p);
1450 assert(offset < UINT64_MAX / secsz);
1451 offset *= secsz;
1452
1453 n_partitions = fdisk_table_get_nents(t);
1454 for (size_t i = 0; i < n_partitions; i++) {
1455 struct fdisk_partition *q;
1456 uint64_t start;
1457
1458 q = fdisk_table_get_partition(t, i);
1459 if (!q)
1460 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to read partition metadata: %m");
1461
1462 if (fdisk_partition_is_used(q) <= 0)
1463 continue;
1464
1465 if (!fdisk_partition_has_start(q))
1466 continue;
1467
1468 start = fdisk_partition_get_start(q);
1469 assert(start < UINT64_MAX / secsz);
1470 start *= secsz;
1471
1472 if (start >= offset && (next == UINT64_MAX || next > start))
1473 next = start;
1474 }
1475
1476 if (next == UINT64_MAX) {
1477 /* No later partition? In that case check the end of the usable area */
1478 next = fdisk_get_last_lba(c);
1479 assert(next < UINT64_MAX);
1480 next++; /* The last LBA is one sector before the end */
1481
1482 assert(next < UINT64_MAX / secsz);
1483 next *= secsz;
1484
1485 if (offset > next)
1486 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Partition end beyond disk end.");
1487 }
1488
1489 assert(next >= offset);
1490 offset = round_up_size(offset, grainsz);
1491 next = round_down_size(next, grainsz);
1492
1493 *ret = LESS_BY(next, offset); /* Saturated subtraction, rounding might have fucked things up */
1494 return 0;
1495 }
1496
1497 static int fdisk_ask_cb(struct fdisk_context *c, struct fdisk_ask *ask, void *data) {
1498 _cleanup_free_ char *ids = NULL;
1499 int r;
1500
1501 if (fdisk_ask_get_type(ask) != FDISK_ASKTYPE_STRING)
1502 return -EINVAL;
1503
1504 ids = new(char, SD_ID128_UUID_STRING_MAX);
1505 if (!ids)
1506 return -ENOMEM;
1507
1508 r = fdisk_ask_string_set_result(ask, sd_id128_to_uuid_string(*(sd_id128_t*) data, ids));
1509 if (r < 0)
1510 return r;
1511
1512 TAKE_PTR(ids);
1513 return 0;
1514 }
1515
1516 static int fdisk_set_disklabel_id_by_uuid(struct fdisk_context *c, sd_id128_t id) {
1517 int r;
1518
1519 r = fdisk_set_ask(c, fdisk_ask_cb, &id);
1520 if (r < 0)
1521 return r;
1522
1523 r = fdisk_set_disklabel_id(c);
1524 if (r < 0)
1525 return r;
1526
1527 return fdisk_set_ask(c, NULL, NULL);
1528 }
1529
1530 static int derive_uuid(sd_id128_t base, const char *token, sd_id128_t *ret) {
1531 union {
1532 uint8_t md[SHA256_DIGEST_SIZE];
1533 sd_id128_t id;
1534 } result;
1535
1536 assert(token);
1537 assert(ret);
1538
1539 /* Derive a new UUID from the specified UUID in a stable and reasonably safe way. Specifically, we
1540 * calculate the HMAC-SHA256 of the specified token string, keyed by the supplied base (typically the
1541 * machine ID). We use the machine ID as key (and not as cleartext!) of the HMAC operation since it's
1542 * the machine ID we don't want to leak. */
1543
1544 hmac_sha256(base.bytes, sizeof(base.bytes), token, strlen(token), result.md);
1545
1546 /* Take the first half, mark it as v4 UUID */
1547 assert_cc(sizeof(result.md) == sizeof(result.id) * 2);
1548 *ret = id128_make_v4_uuid(result.id);
1549 return 0;
1550 }
1551
1552 static int context_load_partition_table(
1553 Context *context,
1554 const char *node,
1555 int *backing_fd) {
1556
1557 _cleanup_(fdisk_unref_contextp) struct fdisk_context *c = NULL;
1558 _cleanup_(fdisk_unref_tablep) struct fdisk_table *t = NULL;
1559 uint64_t left_boundary = UINT64_MAX, first_lba, last_lba, nsectors;
1560 _cleanup_free_ char *disk_uuid_string = NULL;
1561 bool from_scratch = false;
1562 sd_id128_t disk_uuid;
1563 size_t n_partitions;
1564 unsigned long secsz;
1565 uint64_t grainsz;
1566 int r;
1567
1568 assert(context);
1569 assert(node);
1570 assert(backing_fd);
1571 assert(!context->fdisk_context);
1572 assert(!context->free_areas);
1573 assert(context->start == UINT64_MAX);
1574 assert(context->end == UINT64_MAX);
1575 assert(context->total == UINT64_MAX);
1576
1577 c = fdisk_new_context();
1578 if (!c)
1579 return log_oom();
1580
1581 /* libfdisk doesn't have an API to operate on arbitrary fds, hence reopen the fd going via the
1582 * /proc/self/fd/ magic path if we have an existing fd. Open the original file otherwise. */
1583 if (*backing_fd < 0)
1584 r = fdisk_assign_device(c, node, arg_dry_run);
1585 else
1586 r = fdisk_assign_device(c, FORMAT_PROC_FD_PATH(*backing_fd), arg_dry_run);
1587 if (r == -EINVAL && arg_size_auto) {
1588 struct stat st;
1589
1590 /* libfdisk returns EINVAL if opening a file of size zero. Let's check for that, and accept
1591 * it if automatic sizing is requested. */
1592
1593 if (*backing_fd < 0)
1594 r = stat(node, &st);
1595 else
1596 r = fstat(*backing_fd, &st);
1597 if (r < 0)
1598 return log_error_errno(errno, "Failed to stat block device '%s': %m", node);
1599
1600 if (S_ISREG(st.st_mode) && st.st_size == 0) {
1601 /* User the fallback values if we have no better idea */
1602 context->sector_size = 512;
1603 context->grain_size = 4096;
1604 return /* from_scratch = */ true;
1605 }
1606
1607 r = -EINVAL;
1608 }
1609 if (r < 0)
1610 return log_error_errno(r, "Failed to open device '%s': %m", node);
1611
1612 if (*backing_fd < 0) {
1613 /* If we have no fd referencing the device yet, make a copy of the fd now, so that we have one */
1614 *backing_fd = fcntl(fdisk_get_devfd(c), F_DUPFD_CLOEXEC, 3);
1615 if (*backing_fd < 0)
1616 return log_error_errno(errno, "Failed to duplicate fdisk fd: %m");
1617 }
1618
1619 /* Tell udev not to interfere while we are processing the device */
1620 if (flock(fdisk_get_devfd(c), arg_dry_run ? LOCK_SH : LOCK_EX) < 0)
1621 return log_error_errno(errno, "Failed to lock block device: %m");
1622
1623 /* The offsets/sizes libfdisk returns to us will be in multiple of the sector size of the
1624 * device. This is typically 512, and sometimes 4096. Let's query libfdisk once for it, and then use
1625 * it for all our needs. Note that the values we use ourselves always are in bytes though, thus mean
1626 * the same thing universally. Also note that regardless what kind of sector size is in use we'll
1627 * place partitions at multiples of 4K. */
1628 secsz = fdisk_get_sector_size(c);
1629
1630 /* Insist on a power of two, and that it's a multiple of 512, i.e. the traditional sector size. */
1631 if (secsz < 512 || secsz != 1UL << log2u64(secsz))
1632 return log_error_errno(errno, "Sector size %lu is not a power of two larger than 512? Refusing.", secsz);
1633
1634 /* Use at least 4K, and ensure it's a multiple of the sector size, regardless if that is smaller or
1635 * larger */
1636 grainsz = secsz < 4096 ? 4096 : secsz;
1637
1638 log_debug("Sector size of device is %lu bytes. Using grain size of %" PRIu64 ".", secsz, grainsz);
1639
1640 switch (arg_empty) {
1641
1642 case EMPTY_REFUSE:
1643 /* Refuse empty disks, insist on an existing GPT partition table */
1644 if (!fdisk_is_labeltype(c, FDISK_DISKLABEL_GPT))
1645 return log_notice_errno(SYNTHETIC_ERRNO(EHWPOISON), "Disk %s has no GPT disk label, not repartitioning.", node);
1646
1647 break;
1648
1649 case EMPTY_REQUIRE:
1650 /* Require an empty disk, refuse any existing partition table */
1651 r = fdisk_has_label(c);
1652 if (r < 0)
1653 return log_error_errno(r, "Failed to determine whether disk %s has a disk label: %m", node);
1654 if (r > 0)
1655 return log_notice_errno(SYNTHETIC_ERRNO(EHWPOISON), "Disk %s already has a disk label, refusing.", node);
1656
1657 from_scratch = true;
1658 break;
1659
1660 case EMPTY_ALLOW:
1661 /* Allow both an empty disk and an existing partition table, but only GPT */
1662 r = fdisk_has_label(c);
1663 if (r < 0)
1664 return log_error_errno(r, "Failed to determine whether disk %s has a disk label: %m", node);
1665 if (r > 0) {
1666 if (!fdisk_is_labeltype(c, FDISK_DISKLABEL_GPT))
1667 return log_notice_errno(SYNTHETIC_ERRNO(EHWPOISON), "Disk %s has non-GPT disk label, not repartitioning.", node);
1668 } else
1669 from_scratch = true;
1670
1671 break;
1672
1673 case EMPTY_FORCE:
1674 case EMPTY_CREATE:
1675 /* Always reinitiaize the disk, don't consider what there was on the disk before */
1676 from_scratch = true;
1677 break;
1678 }
1679
1680 if (from_scratch) {
1681 r = fdisk_create_disklabel(c, "gpt");
1682 if (r < 0)
1683 return log_error_errno(r, "Failed to create GPT disk label: %m");
1684
1685 r = derive_uuid(context->seed, "disk-uuid", &disk_uuid);
1686 if (r < 0)
1687 return log_error_errno(r, "Failed to acquire disk GPT uuid: %m");
1688
1689 r = fdisk_set_disklabel_id_by_uuid(c, disk_uuid);
1690 if (r < 0)
1691 return log_error_errno(r, "Failed to set GPT disk label: %m");
1692
1693 goto add_initial_free_area;
1694 }
1695
1696 r = fdisk_get_disklabel_id(c, &disk_uuid_string);
1697 if (r < 0)
1698 return log_error_errno(r, "Failed to get current GPT disk label UUID: %m");
1699
1700 r = sd_id128_from_string(disk_uuid_string, &disk_uuid);
1701 if (r < 0)
1702 return log_error_errno(r, "Failed to parse current GPT disk label UUID: %m");
1703
1704 if (sd_id128_is_null(disk_uuid)) {
1705 r = derive_uuid(context->seed, "disk-uuid", &disk_uuid);
1706 if (r < 0)
1707 return log_error_errno(r, "Failed to acquire disk GPT uuid: %m");
1708
1709 r = fdisk_set_disklabel_id(c);
1710 if (r < 0)
1711 return log_error_errno(r, "Failed to set GPT disk label: %m");
1712 }
1713
1714 r = fdisk_get_partitions(c, &t);
1715 if (r < 0)
1716 return log_error_errno(r, "Failed to acquire partition table: %m");
1717
1718 n_partitions = fdisk_table_get_nents(t);
1719 for (size_t i = 0; i < n_partitions; i++) {
1720 _cleanup_free_ char *label_copy = NULL;
1721 Partition *last = NULL;
1722 struct fdisk_partition *p;
1723 struct fdisk_parttype *pt;
1724 const char *pts, *ids, *label;
1725 uint64_t sz, start;
1726 bool found = false;
1727 sd_id128_t ptid, id;
1728 size_t partno;
1729
1730 p = fdisk_table_get_partition(t, i);
1731 if (!p)
1732 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to read partition metadata: %m");
1733
1734 if (fdisk_partition_is_used(p) <= 0)
1735 continue;
1736
1737 if (fdisk_partition_has_start(p) <= 0 ||
1738 fdisk_partition_has_size(p) <= 0 ||
1739 fdisk_partition_has_partno(p) <= 0)
1740 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Found a partition without a position, size or number.");
1741
1742 pt = fdisk_partition_get_type(p);
1743 if (!pt)
1744 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to acquire type of partition: %m");
1745
1746 pts = fdisk_parttype_get_string(pt);
1747 if (!pts)
1748 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to acquire type of partition as string: %m");
1749
1750 r = sd_id128_from_string(pts, &ptid);
1751 if (r < 0)
1752 return log_error_errno(r, "Failed to parse partition type UUID %s: %m", pts);
1753
1754 ids = fdisk_partition_get_uuid(p);
1755 if (!ids)
1756 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Found a partition without a UUID.");
1757
1758 r = sd_id128_from_string(ids, &id);
1759 if (r < 0)
1760 return log_error_errno(r, "Failed to parse partition UUID %s: %m", ids);
1761
1762 label = fdisk_partition_get_name(p);
1763 if (!isempty(label)) {
1764 label_copy = strdup(label);
1765 if (!label_copy)
1766 return log_oom();
1767 }
1768
1769 sz = fdisk_partition_get_size(p);
1770 assert(sz <= UINT64_MAX/secsz);
1771 sz *= secsz;
1772
1773 start = fdisk_partition_get_start(p);
1774 assert(start <= UINT64_MAX/secsz);
1775 start *= secsz;
1776
1777 partno = fdisk_partition_get_partno(p);
1778
1779 if (left_boundary == UINT64_MAX || left_boundary > start)
1780 left_boundary = start;
1781
1782 /* Assign this existing partition to the first partition of the right type that doesn't have
1783 * an existing one assigned yet. */
1784 LIST_FOREACH(partitions, pp, context->partitions) {
1785 last = pp;
1786
1787 if (!sd_id128_equal(pp->type_uuid, ptid))
1788 continue;
1789
1790 if (!pp->current_partition) {
1791 pp->current_uuid = id;
1792 pp->current_size = sz;
1793 pp->offset = start;
1794 pp->partno = partno;
1795 pp->current_label = TAKE_PTR(label_copy);
1796
1797 pp->current_partition = p;
1798 fdisk_ref_partition(p);
1799
1800 r = determine_current_padding(c, t, p, secsz, grainsz, &pp->current_padding);
1801 if (r < 0)
1802 return r;
1803
1804 if (pp->current_padding > 0) {
1805 r = context_add_free_area(context, pp->current_padding, pp);
1806 if (r < 0)
1807 return r;
1808 }
1809
1810 found = true;
1811 break;
1812 }
1813 }
1814
1815 /* If we have no matching definition, create a new one. */
1816 if (!found) {
1817 _cleanup_(partition_freep) Partition *np = NULL;
1818
1819 np = partition_new();
1820 if (!np)
1821 return log_oom();
1822
1823 np->current_uuid = id;
1824 np->type_uuid = ptid;
1825 np->current_size = sz;
1826 np->offset = start;
1827 np->partno = partno;
1828 np->current_label = TAKE_PTR(label_copy);
1829
1830 np->current_partition = p;
1831 fdisk_ref_partition(p);
1832
1833 r = determine_current_padding(c, t, p, secsz, grainsz, &np->current_padding);
1834 if (r < 0)
1835 return r;
1836
1837 if (np->current_padding > 0) {
1838 r = context_add_free_area(context, np->current_padding, np);
1839 if (r < 0)
1840 return r;
1841 }
1842
1843 LIST_INSERT_AFTER(partitions, context->partitions, last, TAKE_PTR(np));
1844 context->n_partitions++;
1845 }
1846 }
1847
1848 add_initial_free_area:
1849 nsectors = fdisk_get_nsectors(c);
1850 assert(nsectors <= UINT64_MAX/secsz);
1851 nsectors *= secsz;
1852
1853 first_lba = fdisk_get_first_lba(c);
1854 assert(first_lba <= UINT64_MAX/secsz);
1855 first_lba *= secsz;
1856
1857 last_lba = fdisk_get_last_lba(c);
1858 assert(last_lba < UINT64_MAX);
1859 last_lba++;
1860 assert(last_lba <= UINT64_MAX/secsz);
1861 last_lba *= secsz;
1862
1863 assert(last_lba >= first_lba);
1864
1865 if (left_boundary == UINT64_MAX) {
1866 /* No partitions at all? Then the whole disk is up for grabs. */
1867
1868 first_lba = round_up_size(first_lba, grainsz);
1869 last_lba = round_down_size(last_lba, grainsz);
1870
1871 if (last_lba > first_lba) {
1872 r = context_add_free_area(context, last_lba - first_lba, NULL);
1873 if (r < 0)
1874 return r;
1875 }
1876 } else {
1877 /* Add space left of first partition */
1878 assert(left_boundary >= first_lba);
1879
1880 first_lba = round_up_size(first_lba, grainsz);
1881 left_boundary = round_down_size(left_boundary, grainsz);
1882 last_lba = round_down_size(last_lba, grainsz);
1883
1884 if (left_boundary > first_lba) {
1885 r = context_add_free_area(context, left_boundary - first_lba, NULL);
1886 if (r < 0)
1887 return r;
1888 }
1889 }
1890
1891 context->start = first_lba;
1892 context->end = last_lba;
1893 context->total = nsectors;
1894 context->sector_size = secsz;
1895 context->grain_size = grainsz;
1896 context->fdisk_context = TAKE_PTR(c);
1897
1898 return from_scratch;
1899 }
1900
1901 static void context_unload_partition_table(Context *context) {
1902 assert(context);
1903
1904 LIST_FOREACH(partitions, p, context->partitions) {
1905
1906 /* Entirely remove partitions that have no configuration */
1907 if (PARTITION_IS_FOREIGN(p)) {
1908 partition_unlink_and_free(context, p);
1909 continue;
1910 }
1911
1912 /* Otherwise drop all data we read off the block device and everything we might have
1913 * calculated based on it */
1914
1915 p->dropped = false;
1916 p->current_size = UINT64_MAX;
1917 p->new_size = UINT64_MAX;
1918 p->current_padding = UINT64_MAX;
1919 p->new_padding = UINT64_MAX;
1920 p->partno = UINT64_MAX;
1921 p->offset = UINT64_MAX;
1922
1923 if (p->current_partition) {
1924 fdisk_unref_partition(p->current_partition);
1925 p->current_partition = NULL;
1926 }
1927
1928 if (p->new_partition) {
1929 fdisk_unref_partition(p->new_partition);
1930 p->new_partition = NULL;
1931 }
1932
1933 p->padding_area = NULL;
1934 p->allocated_to_area = NULL;
1935
1936 p->current_uuid = SD_ID128_NULL;
1937 p->current_label = mfree(p->current_label);
1938 }
1939
1940 context->start = UINT64_MAX;
1941 context->end = UINT64_MAX;
1942 context->total = UINT64_MAX;
1943
1944 if (context->fdisk_context) {
1945 fdisk_unref_context(context->fdisk_context);
1946 context->fdisk_context = NULL;
1947 }
1948
1949 context_free_free_areas(context);
1950 }
1951
1952 static int format_size_change(uint64_t from, uint64_t to, char **ret) {
1953 char *t;
1954
1955 if (from != UINT64_MAX) {
1956 if (from == to || to == UINT64_MAX)
1957 t = strdup(FORMAT_BYTES(from));
1958 else
1959 t = strjoin(FORMAT_BYTES(from), " ", special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), " ", FORMAT_BYTES(to));
1960 } else if (to != UINT64_MAX)
1961 t = strjoin(special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), " ", FORMAT_BYTES(to));
1962 else {
1963 *ret = NULL;
1964 return 0;
1965 }
1966
1967 if (!t)
1968 return log_oom();
1969
1970 *ret = t;
1971 return 1;
1972 }
1973
1974 static const char *partition_label(const Partition *p) {
1975 assert(p);
1976
1977 if (p->new_label)
1978 return p->new_label;
1979
1980 if (p->current_label)
1981 return p->current_label;
1982
1983 return gpt_partition_type_uuid_to_string(p->type_uuid);
1984 }
1985
1986 static int context_dump_partitions(Context *context, const char *node) {
1987 _cleanup_(table_unrefp) Table *t = NULL;
1988 uint64_t sum_padding = 0, sum_size = 0;
1989 int r;
1990
1991 if ((arg_json_format_flags & JSON_FORMAT_OFF) && context->n_partitions == 0) {
1992 log_info("Empty partition table.");
1993 return 0;
1994 }
1995
1996 t = table_new("type", "label", "uuid", "file", "node", "offset", "old size", "raw size", "size", "old padding", "raw padding", "padding", "activity");
1997 if (!t)
1998 return log_oom();
1999
2000 if (!DEBUG_LOGGING) {
2001 if (arg_json_format_flags & JSON_FORMAT_OFF)
2002 (void) table_set_display(t, (size_t) 0, (size_t) 1, (size_t) 2, (size_t) 3, (size_t) 4,
2003 (size_t) 8, (size_t) 11);
2004 else
2005 (void) table_set_display(t, (size_t) 0, (size_t) 1, (size_t) 2, (size_t) 3, (size_t) 4,
2006 (size_t) 5, (size_t) 6, (size_t) 7, (size_t) 9, (size_t) 10, (size_t) 12);
2007 }
2008
2009 (void) table_set_align_percent(t, table_get_cell(t, 0, 5), 100);
2010 (void) table_set_align_percent(t, table_get_cell(t, 0, 6), 100);
2011 (void) table_set_align_percent(t, table_get_cell(t, 0, 7), 100);
2012 (void) table_set_align_percent(t, table_get_cell(t, 0, 8), 100);
2013 (void) table_set_align_percent(t, table_get_cell(t, 0, 9), 100);
2014 (void) table_set_align_percent(t, table_get_cell(t, 0, 10), 100);
2015 (void) table_set_align_percent(t, table_get_cell(t, 0, 11), 100);
2016
2017 LIST_FOREACH(partitions, p, context->partitions) {
2018 _cleanup_free_ char *size_change = NULL, *padding_change = NULL, *partname = NULL;
2019 char uuid_buffer[SD_ID128_UUID_STRING_MAX];
2020 const char *label, *activity = NULL;
2021
2022 if (p->dropped)
2023 continue;
2024
2025 if (p->current_size == UINT64_MAX)
2026 activity = "create";
2027 else if (p->current_size != p->new_size)
2028 activity = "resize";
2029
2030 label = partition_label(p);
2031 partname = p->partno != UINT64_MAX ? fdisk_partname(node, p->partno+1) : NULL;
2032
2033 r = format_size_change(p->current_size, p->new_size, &size_change);
2034 if (r < 0)
2035 return r;
2036
2037 r = format_size_change(p->current_padding, p->new_padding, &padding_change);
2038 if (r < 0)
2039 return r;
2040
2041 if (p->new_size != UINT64_MAX)
2042 sum_size += p->new_size;
2043 if (p->new_padding != UINT64_MAX)
2044 sum_padding += p->new_padding;
2045
2046 r = table_add_many(
2047 t,
2048 TABLE_STRING, gpt_partition_type_uuid_to_string_harder(p->type_uuid, uuid_buffer),
2049 TABLE_STRING, empty_to_null(label) ?: "-", TABLE_SET_COLOR, empty_to_null(label) ? NULL : ansi_grey(),
2050 TABLE_UUID, sd_id128_is_null(p->new_uuid) ? p->current_uuid : p->new_uuid,
2051 TABLE_STRING, p->definition_path ? basename(p->definition_path) : "-", TABLE_SET_COLOR, p->definition_path ? NULL : ansi_grey(),
2052 TABLE_STRING, partname ?: "-", TABLE_SET_COLOR, partname ? NULL : ansi_highlight(),
2053 TABLE_UINT64, p->offset,
2054 TABLE_UINT64, p->current_size == UINT64_MAX ? 0 : p->current_size,
2055 TABLE_UINT64, p->new_size,
2056 TABLE_STRING, size_change, TABLE_SET_COLOR, !p->partitions_next && sum_size > 0 ? ansi_underline() : NULL,
2057 TABLE_UINT64, p->current_padding == UINT64_MAX ? 0 : p->current_padding,
2058 TABLE_UINT64, p->new_padding,
2059 TABLE_STRING, padding_change, TABLE_SET_COLOR, !p->partitions_next && sum_padding > 0 ? ansi_underline() : NULL,
2060 TABLE_STRING, activity ?: "unchanged");
2061 if (r < 0)
2062 return table_log_add_error(r);
2063 }
2064
2065 if ((arg_json_format_flags & JSON_FORMAT_OFF) && (sum_padding > 0 || sum_size > 0)) {
2066 const char *a, *b;
2067
2068 a = strjoina(special_glyph(SPECIAL_GLYPH_SIGMA), " = ", FORMAT_BYTES(sum_size));
2069 b = strjoina(special_glyph(SPECIAL_GLYPH_SIGMA), " = ", FORMAT_BYTES(sum_padding));
2070
2071 r = table_add_many(
2072 t,
2073 TABLE_EMPTY,
2074 TABLE_EMPTY,
2075 TABLE_EMPTY,
2076 TABLE_EMPTY,
2077 TABLE_EMPTY,
2078 TABLE_EMPTY,
2079 TABLE_EMPTY,
2080 TABLE_EMPTY,
2081 TABLE_STRING, a,
2082 TABLE_EMPTY,
2083 TABLE_EMPTY,
2084 TABLE_STRING, b,
2085 TABLE_EMPTY);
2086 if (r < 0)
2087 return table_log_add_error(r);
2088 }
2089
2090 return table_print_with_pager(t, arg_json_format_flags, arg_pager_flags, arg_legend);
2091 }
2092
2093 static void context_bar_char_process_partition(
2094 Context *context,
2095 Partition *bar[],
2096 size_t n,
2097 Partition *p,
2098 size_t *ret_start) {
2099
2100 uint64_t from, to, total;
2101 size_t x, y;
2102
2103 assert(context);
2104 assert(bar);
2105 assert(n > 0);
2106 assert(p);
2107
2108 if (p->dropped)
2109 return;
2110
2111 assert(p->offset != UINT64_MAX);
2112 assert(p->new_size != UINT64_MAX);
2113
2114 from = p->offset;
2115 to = from + p->new_size;
2116
2117 assert(context->total > 0);
2118 total = context->total;
2119
2120 assert(from <= total);
2121 x = from * n / total;
2122
2123 assert(to <= total);
2124 y = to * n / total;
2125
2126 assert(x <= y);
2127 assert(y <= n);
2128
2129 for (size_t i = x; i < y; i++)
2130 bar[i] = p;
2131
2132 *ret_start = x;
2133 }
2134
2135 static int partition_hint(const Partition *p, const char *node, char **ret) {
2136 _cleanup_free_ char *buf = NULL;
2137 const char *label;
2138 sd_id128_t id;
2139
2140 /* Tries really hard to find a suitable description for this partition */
2141
2142 if (p->definition_path) {
2143 buf = strdup(basename(p->definition_path));
2144 goto done;
2145 }
2146
2147 label = partition_label(p);
2148 if (!isempty(label)) {
2149 buf = strdup(label);
2150 goto done;
2151 }
2152
2153 if (p->partno != UINT64_MAX) {
2154 buf = fdisk_partname(node, p->partno+1);
2155 goto done;
2156 }
2157
2158 if (!sd_id128_is_null(p->new_uuid))
2159 id = p->new_uuid;
2160 else if (!sd_id128_is_null(p->current_uuid))
2161 id = p->current_uuid;
2162 else
2163 id = p->type_uuid;
2164
2165 buf = strdup(SD_ID128_TO_UUID_STRING(id));
2166
2167 done:
2168 if (!buf)
2169 return -ENOMEM;
2170
2171 *ret = TAKE_PTR(buf);
2172 return 0;
2173 }
2174
2175 static int context_dump_partition_bar(Context *context, const char *node) {
2176 _cleanup_free_ Partition **bar = NULL;
2177 _cleanup_free_ size_t *start_array = NULL;
2178 Partition *last = NULL;
2179 bool z = false;
2180 size_t c, j = 0;
2181
2182 assert_se((c = columns()) >= 2);
2183 c -= 2; /* We do not use the leftmost and rightmost character cell */
2184
2185 bar = new0(Partition*, c);
2186 if (!bar)
2187 return log_oom();
2188
2189 start_array = new(size_t, context->n_partitions);
2190 if (!start_array)
2191 return log_oom();
2192
2193 LIST_FOREACH(partitions, p, context->partitions)
2194 context_bar_char_process_partition(context, bar, c, p, start_array + j++);
2195
2196 putc(' ', stdout);
2197
2198 for (size_t i = 0; i < c; i++) {
2199 if (bar[i]) {
2200 if (last != bar[i])
2201 z = !z;
2202
2203 fputs(z ? ansi_green() : ansi_yellow(), stdout);
2204 fputs(special_glyph(SPECIAL_GLYPH_DARK_SHADE), stdout);
2205 } else {
2206 fputs(ansi_normal(), stdout);
2207 fputs(special_glyph(SPECIAL_GLYPH_LIGHT_SHADE), stdout);
2208 }
2209
2210 last = bar[i];
2211 }
2212
2213 fputs(ansi_normal(), stdout);
2214 putc('\n', stdout);
2215
2216 for (size_t i = 0; i < context->n_partitions; i++) {
2217 _cleanup_free_ char **line = NULL;
2218
2219 line = new0(char*, c);
2220 if (!line)
2221 return log_oom();
2222
2223 j = 0;
2224 LIST_FOREACH(partitions, p, context->partitions) {
2225 _cleanup_free_ char *d = NULL;
2226 j++;
2227
2228 if (i < context->n_partitions - j) {
2229
2230 if (line[start_array[j-1]]) {
2231 const char *e;
2232
2233 /* Upgrade final corner to the right with a branch to the right */
2234 e = startswith(line[start_array[j-1]], special_glyph(SPECIAL_GLYPH_TREE_RIGHT));
2235 if (e) {
2236 d = strjoin(special_glyph(SPECIAL_GLYPH_TREE_BRANCH), e);
2237 if (!d)
2238 return log_oom();
2239 }
2240 }
2241
2242 if (!d) {
2243 d = strdup(special_glyph(SPECIAL_GLYPH_TREE_VERTICAL));
2244 if (!d)
2245 return log_oom();
2246 }
2247
2248 } else if (i == context->n_partitions - j) {
2249 _cleanup_free_ char *hint = NULL;
2250
2251 (void) partition_hint(p, node, &hint);
2252
2253 if (streq_ptr(line[start_array[j-1]], special_glyph(SPECIAL_GLYPH_TREE_VERTICAL)))
2254 d = strjoin(special_glyph(SPECIAL_GLYPH_TREE_BRANCH), " ", strna(hint));
2255 else
2256 d = strjoin(special_glyph(SPECIAL_GLYPH_TREE_RIGHT), " ", strna(hint));
2257
2258 if (!d)
2259 return log_oom();
2260 }
2261
2262 if (d)
2263 free_and_replace(line[start_array[j-1]], d);
2264 }
2265
2266 putc(' ', stdout);
2267
2268 j = 0;
2269 while (j < c) {
2270 if (line[j]) {
2271 fputs(line[j], stdout);
2272 j += utf8_console_width(line[j]);
2273 } else {
2274 putc(' ', stdout);
2275 j++;
2276 }
2277 }
2278
2279 putc('\n', stdout);
2280
2281 for (j = 0; j < c; j++)
2282 free(line[j]);
2283 }
2284
2285 return 0;
2286 }
2287
2288 static bool context_changed(const Context *context) {
2289 assert(context);
2290
2291 LIST_FOREACH(partitions, p, context->partitions) {
2292 if (p->dropped)
2293 continue;
2294
2295 if (p->allocated_to_area)
2296 return true;
2297
2298 if (p->new_size != p->current_size)
2299 return true;
2300 }
2301
2302 return false;
2303 }
2304
2305 static int context_wipe_range(Context *context, uint64_t offset, uint64_t size) {
2306 _cleanup_(blkid_free_probep) blkid_probe probe = NULL;
2307 int r;
2308
2309 assert(context);
2310 assert(offset != UINT64_MAX);
2311 assert(size != UINT64_MAX);
2312
2313 probe = blkid_new_probe();
2314 if (!probe)
2315 return log_oom();
2316
2317 errno = 0;
2318 r = blkid_probe_set_device(probe, fdisk_get_devfd(context->fdisk_context), offset, size);
2319 if (r < 0)
2320 return log_error_errno(errno ?: SYNTHETIC_ERRNO(EIO), "Failed to allocate device probe for wiping.");
2321
2322 errno = 0;
2323 if (blkid_probe_enable_superblocks(probe, true) < 0 ||
2324 blkid_probe_set_superblocks_flags(probe, BLKID_SUBLKS_MAGIC|BLKID_SUBLKS_BADCSUM) < 0 ||
2325 blkid_probe_enable_partitions(probe, true) < 0 ||
2326 blkid_probe_set_partitions_flags(probe, BLKID_PARTS_MAGIC) < 0)
2327 return log_error_errno(errno ?: SYNTHETIC_ERRNO(EIO), "Failed to enable superblock and partition probing.");
2328
2329 for (;;) {
2330 errno = 0;
2331 r = blkid_do_probe(probe);
2332 if (r < 0)
2333 return log_error_errno(errno ?: SYNTHETIC_ERRNO(EIO), "Failed to probe for file systems.");
2334 if (r > 0)
2335 break;
2336
2337 errno = 0;
2338 if (blkid_do_wipe(probe, false) < 0)
2339 return log_error_errno(errno ?: SYNTHETIC_ERRNO(EIO), "Failed to wipe file system signature.");
2340 }
2341
2342 return 0;
2343 }
2344
2345 static int context_wipe_partition(Context *context, Partition *p) {
2346 int r;
2347
2348 assert(context);
2349 assert(p);
2350 assert(!PARTITION_EXISTS(p)); /* Safety check: never wipe existing partitions */
2351
2352 assert(p->offset != UINT64_MAX);
2353 assert(p->new_size != UINT64_MAX);
2354
2355 r = context_wipe_range(context, p->offset, p->new_size);
2356 if (r < 0)
2357 return r;
2358
2359 log_info("Successfully wiped file system signatures from future partition %" PRIu64 ".", p->partno);
2360 return 0;
2361 }
2362
2363 static int context_discard_range(
2364 Context *context,
2365 uint64_t offset,
2366 uint64_t size) {
2367
2368 struct stat st;
2369 int fd;
2370
2371 assert(context);
2372 assert(offset != UINT64_MAX);
2373 assert(size != UINT64_MAX);
2374
2375 if (size <= 0)
2376 return 0;
2377
2378 assert_se((fd = fdisk_get_devfd(context->fdisk_context)) >= 0);
2379
2380 if (fstat(fd, &st) < 0)
2381 return -errno;
2382
2383 if (S_ISREG(st.st_mode)) {
2384 if (fallocate(fd, FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE, offset, size) < 0) {
2385 if (ERRNO_IS_NOT_SUPPORTED(errno))
2386 return -EOPNOTSUPP;
2387
2388 return -errno;
2389 }
2390
2391 return 1;
2392 }
2393
2394 if (S_ISBLK(st.st_mode)) {
2395 uint64_t range[2], end;
2396
2397 range[0] = round_up_size(offset, context->sector_size);
2398
2399 if (offset > UINT64_MAX - size)
2400 return -ERANGE;
2401
2402 end = offset + size;
2403 if (end <= range[0])
2404 return 0;
2405
2406 range[1] = round_down_size(end - range[0], context->sector_size);
2407 if (range[1] <= 0)
2408 return 0;
2409
2410 if (ioctl(fd, BLKDISCARD, range) < 0) {
2411 if (ERRNO_IS_NOT_SUPPORTED(errno))
2412 return -EOPNOTSUPP;
2413
2414 return -errno;
2415 }
2416
2417 return 1;
2418 }
2419
2420 return -EOPNOTSUPP;
2421 }
2422
2423 static int context_discard_partition(Context *context, Partition *p) {
2424 int r;
2425
2426 assert(context);
2427 assert(p);
2428
2429 assert(p->offset != UINT64_MAX);
2430 assert(p->new_size != UINT64_MAX);
2431 assert(!PARTITION_EXISTS(p)); /* Safety check: never discard existing partitions */
2432
2433 if (!arg_discard)
2434 return 0;
2435
2436 r = context_discard_range(context, p->offset, p->new_size);
2437 if (r == -EOPNOTSUPP) {
2438 log_info("Storage does not support discard, not discarding data in future partition %" PRIu64 ".", p->partno);
2439 return 0;
2440 }
2441 if (r == -EBUSY) {
2442 /* Let's handle this gracefully: https://bugzilla.kernel.org/show_bug.cgi?id=211167 */
2443 log_info("Block device is busy, not discarding partition %" PRIu64 " because it probably is mounted.", p->partno);
2444 return 0;
2445 }
2446 if (r == 0) {
2447 log_info("Partition %" PRIu64 " too short for discard, skipping.", p->partno);
2448 return 0;
2449 }
2450 if (r < 0)
2451 return log_error_errno(r, "Failed to discard data for future partition %" PRIu64 ".", p->partno);
2452
2453 log_info("Successfully discarded data from future partition %" PRIu64 ".", p->partno);
2454 return 1;
2455 }
2456
2457 static int context_discard_gap_after(Context *context, Partition *p) {
2458 uint64_t gap, next = UINT64_MAX;
2459 int r;
2460
2461 assert(context);
2462 assert(!p || (p->offset != UINT64_MAX && p->new_size != UINT64_MAX));
2463
2464 if (p)
2465 gap = p->offset + p->new_size;
2466 else
2467 gap = context->start;
2468
2469 LIST_FOREACH(partitions, q, context->partitions) {
2470 if (q->dropped)
2471 continue;
2472
2473 assert(q->offset != UINT64_MAX);
2474 assert(q->new_size != UINT64_MAX);
2475
2476 if (q->offset < gap)
2477 continue;
2478
2479 if (next == UINT64_MAX || q->offset < next)
2480 next = q->offset;
2481 }
2482
2483 if (next == UINT64_MAX) {
2484 next = context->end;
2485 if (gap > next)
2486 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Partition end beyond disk end.");
2487 }
2488
2489 assert(next >= gap);
2490 r = context_discard_range(context, gap, next - gap);
2491 if (r == -EOPNOTSUPP) {
2492 if (p)
2493 log_info("Storage does not support discard, not discarding gap after partition %" PRIu64 ".", p->partno);
2494 else
2495 log_info("Storage does not support discard, not discarding gap at beginning of disk.");
2496 return 0;
2497 }
2498 if (r == 0) /* Too short */
2499 return 0;
2500 if (r < 0) {
2501 if (p)
2502 return log_error_errno(r, "Failed to discard gap after partition %" PRIu64 ".", p->partno);
2503 else
2504 return log_error_errno(r, "Failed to discard gap at beginning of disk.");
2505 }
2506
2507 if (p)
2508 log_info("Successfully discarded gap after partition %" PRIu64 ".", p->partno);
2509 else
2510 log_info("Successfully discarded gap at beginning of disk.");
2511
2512 return 0;
2513 }
2514
2515 static int context_wipe_and_discard(Context *context, bool from_scratch) {
2516 int r;
2517
2518 assert(context);
2519
2520 /* Wipe and discard the contents of all partitions we are about to create. We skip the discarding if
2521 * we were supposed to start from scratch anyway, as in that case we just discard the whole block
2522 * device in one go early on. */
2523
2524 LIST_FOREACH(partitions, p, context->partitions) {
2525
2526 if (!p->allocated_to_area)
2527 continue;
2528
2529 r = context_wipe_partition(context, p);
2530 if (r < 0)
2531 return r;
2532
2533 if (!from_scratch) {
2534 r = context_discard_partition(context, p);
2535 if (r < 0)
2536 return r;
2537
2538 r = context_discard_gap_after(context, p);
2539 if (r < 0)
2540 return r;
2541 }
2542 }
2543
2544 if (!from_scratch) {
2545 r = context_discard_gap_after(context, NULL);
2546 if (r < 0)
2547 return r;
2548 }
2549
2550 return 0;
2551 }
2552
2553 static int partition_encrypt(
2554 Context *context,
2555 Partition *p,
2556 const char *node,
2557 struct crypt_device **ret_cd,
2558 char **ret_volume,
2559 int *ret_fd) {
2560 #if HAVE_LIBCRYPTSETUP
2561 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
2562 _cleanup_(erase_and_freep) void *volume_key = NULL;
2563 _cleanup_free_ char *dm_name = NULL, *vol = NULL;
2564 size_t volume_key_size = 256 / 8;
2565 sd_id128_t uuid;
2566 int r;
2567
2568 assert(context);
2569 assert(p);
2570 assert(p->encrypt != ENCRYPT_OFF);
2571
2572 log_debug("Encryption mode for partition %" PRIu64 ": %s", p->partno, encrypt_mode_to_string(p->encrypt));
2573
2574 r = dlopen_cryptsetup();
2575 if (r < 0)
2576 return log_error_errno(r, "libcryptsetup not found, cannot encrypt: %m");
2577
2578 if (asprintf(&dm_name, "luks-repart-%08" PRIx64, random_u64()) < 0)
2579 return log_oom();
2580
2581 if (ret_volume) {
2582 vol = path_join("/dev/mapper/", dm_name);
2583 if (!vol)
2584 return log_oom();
2585 }
2586
2587 r = derive_uuid(p->new_uuid, "luks-uuid", &uuid);
2588 if (r < 0)
2589 return r;
2590
2591 log_info("Encrypting future partition %" PRIu64 "...", p->partno);
2592
2593 volume_key = malloc(volume_key_size);
2594 if (!volume_key)
2595 return log_oom();
2596
2597 r = crypto_random_bytes(volume_key, volume_key_size);
2598 if (r < 0)
2599 return log_error_errno(r, "Failed to generate volume key: %m");
2600
2601 r = sym_crypt_init(&cd, node);
2602 if (r < 0)
2603 return log_error_errno(r, "Failed to allocate libcryptsetup context: %m");
2604
2605 cryptsetup_enable_logging(cd);
2606
2607 r = sym_crypt_format(cd,
2608 CRYPT_LUKS2,
2609 "aes",
2610 "xts-plain64",
2611 SD_ID128_TO_UUID_STRING(uuid),
2612 volume_key,
2613 volume_key_size,
2614 &(struct crypt_params_luks2) {
2615 .label = strempty(p->new_label),
2616 .sector_size = context->sector_size,
2617 });
2618 if (r < 0)
2619 return log_error_errno(r, "Failed to LUKS2 format future partition: %m");
2620
2621 if (IN_SET(p->encrypt, ENCRYPT_KEY_FILE, ENCRYPT_KEY_FILE_TPM2)) {
2622 r = sym_crypt_keyslot_add_by_volume_key(
2623 cd,
2624 CRYPT_ANY_SLOT,
2625 volume_key,
2626 volume_key_size,
2627 strempty(arg_key),
2628 arg_key_size);
2629 if (r < 0)
2630 return log_error_errno(r, "Failed to add LUKS2 key: %m");
2631 }
2632
2633 if (IN_SET(p->encrypt, ENCRYPT_TPM2, ENCRYPT_KEY_FILE_TPM2)) {
2634 #if HAVE_TPM2
2635 _cleanup_(erase_and_freep) char *base64_encoded = NULL;
2636 _cleanup_(json_variant_unrefp) JsonVariant *v = NULL;
2637 _cleanup_(erase_and_freep) void *secret = NULL;
2638 _cleanup_free_ void *blob = NULL, *hash = NULL;
2639 size_t secret_size, blob_size, hash_size;
2640 uint16_t pcr_bank, primary_alg;
2641 int keyslot;
2642
2643 r = tpm2_seal(arg_tpm2_device, arg_tpm2_pcr_mask, NULL, &secret, &secret_size, &blob, &blob_size, &hash, &hash_size, &pcr_bank, &primary_alg);
2644 if (r < 0)
2645 return log_error_errno(r, "Failed to seal to TPM2: %m");
2646
2647 r = base64mem(secret, secret_size, &base64_encoded);
2648 if (r < 0)
2649 return log_error_errno(r, "Failed to base64 encode secret key: %m");
2650
2651 r = cryptsetup_set_minimal_pbkdf(cd);
2652 if (r < 0)
2653 return log_error_errno(r, "Failed to set minimal PBKDF: %m");
2654
2655 keyslot = sym_crypt_keyslot_add_by_volume_key(
2656 cd,
2657 CRYPT_ANY_SLOT,
2658 volume_key,
2659 volume_key_size,
2660 base64_encoded,
2661 strlen(base64_encoded));
2662 if (keyslot < 0)
2663 return log_error_errno(keyslot, "Failed to add new TPM2 key to %s: %m", node);
2664
2665 r = tpm2_make_luks2_json(keyslot, arg_tpm2_pcr_mask, pcr_bank, primary_alg, blob, blob_size, hash, hash_size, 0, &v);
2666 if (r < 0)
2667 return log_error_errno(r, "Failed to prepare TPM2 JSON token object: %m");
2668
2669 r = cryptsetup_add_token_json(cd, v);
2670 if (r < 0)
2671 return log_error_errno(r, "Failed to add TPM2 JSON token to LUKS2 header: %m");
2672 #else
2673 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
2674 "Support for TPM2 enrollment not enabled.");
2675 #endif
2676 }
2677
2678 r = sym_crypt_activate_by_volume_key(
2679 cd,
2680 dm_name,
2681 volume_key,
2682 volume_key_size,
2683 arg_discard ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0);
2684 if (r < 0)
2685 return log_error_errno(r, "Failed to activate LUKS superblock: %m");
2686
2687 log_info("Successfully encrypted future partition %" PRIu64 ".", p->partno);
2688
2689 if (ret_fd) {
2690 _cleanup_close_ int dev_fd = -1;
2691
2692 dev_fd = open(vol, O_RDWR|O_CLOEXEC|O_NOCTTY);
2693 if (dev_fd < 0)
2694 return log_error_errno(errno, "Failed to open LUKS volume '%s': %m", vol);
2695
2696 *ret_fd = TAKE_FD(dev_fd);
2697 }
2698
2699 if (ret_cd)
2700 *ret_cd = TAKE_PTR(cd);
2701 if (ret_volume)
2702 *ret_volume = TAKE_PTR(vol);
2703
2704 return 0;
2705 #else
2706 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "libcryptsetup is not supported, cannot encrypt: %m");
2707 #endif
2708 }
2709
2710 static int deactivate_luks(struct crypt_device *cd, const char *node) {
2711 #if HAVE_LIBCRYPTSETUP
2712 int r;
2713
2714 if (!cd)
2715 return 0;
2716
2717 assert(node);
2718
2719 /* udev or so might access out block device in the background while we are done. Let's hence force
2720 * detach the volume. We sync'ed before, hence this should be safe. */
2721
2722 r = sym_crypt_deactivate_by_name(cd, basename(node), CRYPT_DEACTIVATE_FORCE);
2723 if (r < 0)
2724 return log_error_errno(r, "Failed to deactivate LUKS device: %m");
2725
2726 return 1;
2727 #else
2728 return 0;
2729 #endif
2730 }
2731
2732 static int context_copy_blocks(Context *context) {
2733 int whole_fd = -1, r;
2734
2735 assert(context);
2736
2737 /* Copy in file systems on the block level */
2738
2739 LIST_FOREACH(partitions, p, context->partitions) {
2740 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
2741 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
2742 _cleanup_free_ char *encrypted = NULL;
2743 _cleanup_close_ int encrypted_dev_fd = -1;
2744 int target_fd;
2745
2746 if (p->copy_blocks_fd < 0)
2747 continue;
2748
2749 if (p->dropped)
2750 continue;
2751
2752 if (PARTITION_EXISTS(p)) /* Never copy over existing partitions */
2753 continue;
2754
2755 assert(p->new_size != UINT64_MAX);
2756 assert(p->copy_blocks_size != UINT64_MAX);
2757 assert(p->new_size >= p->copy_blocks_size);
2758
2759 if (whole_fd < 0)
2760 assert_se((whole_fd = fdisk_get_devfd(context->fdisk_context)) >= 0);
2761
2762 if (p->encrypt != ENCRYPT_OFF) {
2763 r = loop_device_make(whole_fd, O_RDWR, p->offset, p->new_size, 0, &d);
2764 if (r < 0)
2765 return log_error_errno(r, "Failed to make loopback device of future partition %" PRIu64 ": %m", p->partno);
2766
2767 r = loop_device_flock(d, LOCK_EX);
2768 if (r < 0)
2769 return log_error_errno(r, "Failed to lock loopback device: %m");
2770
2771 r = partition_encrypt(context, p, d->node, &cd, &encrypted, &encrypted_dev_fd);
2772 if (r < 0)
2773 return log_error_errno(r, "Failed to encrypt device: %m");
2774
2775 if (flock(encrypted_dev_fd, LOCK_EX) < 0)
2776 return log_error_errno(errno, "Failed to lock LUKS device: %m");
2777
2778 target_fd = encrypted_dev_fd;
2779 } else {
2780 if (lseek(whole_fd, p->offset, SEEK_SET) == (off_t) -1)
2781 return log_error_errno(errno, "Failed to seek to partition offset: %m");
2782
2783 target_fd = whole_fd;
2784 }
2785
2786 log_info("Copying in '%s' (%s) on block level into future partition %" PRIu64 ".",
2787 p->copy_blocks_path, FORMAT_BYTES(p->copy_blocks_size), p->partno);
2788
2789 r = copy_bytes_full(p->copy_blocks_fd, target_fd, p->copy_blocks_size, 0, NULL, NULL, NULL, NULL);
2790 if (r < 0)
2791 return log_error_errno(r, "Failed to copy in data from '%s': %m", p->copy_blocks_path);
2792
2793 if (fsync(target_fd) < 0)
2794 return log_error_errno(errno, "Failed to synchronize copied data blocks: %m");
2795
2796 if (p->encrypt != ENCRYPT_OFF) {
2797 encrypted_dev_fd = safe_close(encrypted_dev_fd);
2798
2799 r = deactivate_luks(cd, encrypted);
2800 if (r < 0)
2801 return r;
2802
2803 sym_crypt_free(cd);
2804 cd = NULL;
2805
2806 r = loop_device_sync(d);
2807 if (r < 0)
2808 return log_error_errno(r, "Failed to sync loopback device: %m");
2809 }
2810
2811 log_info("Copying in of '%s' on block level completed.", p->copy_blocks_path);
2812 }
2813
2814 return 0;
2815 }
2816
2817 static int do_copy_files(Partition *p, const char *fs) {
2818 int r;
2819
2820 assert(p);
2821 assert(fs);
2822
2823 STRV_FOREACH_PAIR(source, target, p->copy_files) {
2824 _cleanup_close_ int sfd = -1, pfd = -1, tfd = -1;
2825
2826 sfd = chase_symlinks_and_open(*source, arg_root, CHASE_PREFIX_ROOT|CHASE_WARN, O_CLOEXEC|O_NOCTTY, NULL);
2827 if (sfd < 0)
2828 return log_error_errno(sfd, "Failed to open source file '%s%s': %m", strempty(arg_root), *source);
2829
2830 r = fd_verify_regular(sfd);
2831 if (r < 0) {
2832 if (r != -EISDIR)
2833 return log_error_errno(r, "Failed to check type of source file '%s': %m", *source);
2834
2835 /* We are looking at a directory */
2836 tfd = chase_symlinks_and_open(*target, fs, CHASE_PREFIX_ROOT|CHASE_WARN, O_RDONLY|O_DIRECTORY|O_CLOEXEC, NULL);
2837 if (tfd < 0) {
2838 _cleanup_free_ char *dn = NULL, *fn = NULL;
2839
2840 if (tfd != -ENOENT)
2841 return log_error_errno(tfd, "Failed to open target directory '%s': %m", *target);
2842
2843 r = path_extract_filename(*target, &fn);
2844 if (r < 0)
2845 return log_error_errno(r, "Failed to extract filename from '%s': %m", *target);
2846
2847 r = path_extract_directory(*target, &dn);
2848 if (r < 0)
2849 return log_error_errno(r, "Failed to extract directory from '%s': %m", *target);
2850
2851 r = mkdir_p_root(fs, dn, UID_INVALID, GID_INVALID, 0755);
2852 if (r < 0)
2853 return log_error_errno(r, "Failed to create parent directory '%s': %m", dn);
2854
2855 pfd = chase_symlinks_and_open(dn, fs, CHASE_PREFIX_ROOT|CHASE_WARN, O_RDONLY|O_DIRECTORY|O_CLOEXEC, NULL);
2856 if (pfd < 0)
2857 return log_error_errno(pfd, "Failed to open parent directory of target: %m");
2858
2859 r = copy_tree_at(
2860 sfd, ".",
2861 pfd, fn,
2862 UID_INVALID, GID_INVALID,
2863 COPY_REFLINK|COPY_MERGE|COPY_REPLACE|COPY_SIGINT|COPY_HARDLINKS|COPY_ALL_XATTRS);
2864 } else
2865 r = copy_tree_at(
2866 sfd, ".",
2867 tfd, ".",
2868 UID_INVALID, GID_INVALID,
2869 COPY_REFLINK|COPY_MERGE|COPY_REPLACE|COPY_SIGINT|COPY_HARDLINKS|COPY_ALL_XATTRS);
2870 if (r < 0)
2871 return log_error_errno(r, "Failed to copy '%s' to '%s%s': %m", *source, strempty(arg_root), *target);
2872 } else {
2873 _cleanup_free_ char *dn = NULL, *fn = NULL;
2874
2875 /* We are looking at a regular file */
2876
2877 r = path_extract_filename(*target, &fn);
2878 if (r == -EADDRNOTAVAIL || r == O_DIRECTORY)
2879 return log_error_errno(SYNTHETIC_ERRNO(EISDIR),
2880 "Target path '%s' refers to a directory, but source path '%s' refers to regular file, can't copy.", *target, *source);
2881 if (r < 0)
2882 return log_error_errno(r, "Failed to extract filename from '%s': %m", *target);
2883
2884 r = path_extract_directory(*target, &dn);
2885 if (r < 0)
2886 return log_error_errno(r, "Failed to extract directory from '%s': %m", *target);
2887
2888 r = mkdir_p_root(fs, dn, UID_INVALID, GID_INVALID, 0755);
2889 if (r < 0)
2890 return log_error_errno(r, "Failed to create parent directory: %m");
2891
2892 pfd = chase_symlinks_and_open(dn, fs, CHASE_PREFIX_ROOT|CHASE_WARN, O_RDONLY|O_DIRECTORY|O_CLOEXEC, NULL);
2893 if (pfd < 0)
2894 return log_error_errno(pfd, "Failed to open parent directory of target: %m");
2895
2896 tfd = openat(pfd, fn, O_CREAT|O_EXCL|O_WRONLY|O_CLOEXEC, 0700);
2897 if (tfd < 0)
2898 return log_error_errno(errno, "Failed to create target file '%s': %m", *target);
2899
2900 r = copy_bytes(sfd, tfd, UINT64_MAX, COPY_REFLINK|COPY_SIGINT);
2901 if (r < 0)
2902 return log_error_errno(r, "Failed to copy '%s' to '%s%s': %m", *source, strempty(arg_root), *target);
2903
2904 (void) copy_xattr(sfd, tfd, COPY_ALL_XATTRS);
2905 (void) copy_access(sfd, tfd);
2906 (void) copy_times(sfd, tfd, 0);
2907 }
2908 }
2909
2910 return 0;
2911 }
2912
2913 static int do_make_directories(Partition *p, const char *fs) {
2914 int r;
2915
2916 assert(p);
2917 assert(fs);
2918
2919 STRV_FOREACH(d, p->make_directories) {
2920
2921 r = mkdir_p_root(fs, *d, UID_INVALID, GID_INVALID, 0755);
2922 if (r < 0)
2923 return log_error_errno(r, "Failed to create directory '%s' in file system: %m", *d);
2924 }
2925
2926 return 0;
2927 }
2928
2929 static int partition_populate(Partition *p, const char *node) {
2930 int r;
2931
2932 assert(p);
2933 assert(node);
2934
2935 if (strv_isempty(p->copy_files) && strv_isempty(p->make_directories))
2936 return 0;
2937
2938 log_info("Populating partition %" PRIu64 " with files.", p->partno);
2939
2940 /* We copy in a child process, since we have to mount the fs for that, and we don't want that fs to
2941 * appear in the host namespace. Hence we fork a child that has its own file system namespace and
2942 * detached mount propagation. */
2943
2944 r = safe_fork("(sd-copy)", FORK_DEATHSIG|FORK_LOG|FORK_WAIT|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, NULL);
2945 if (r < 0)
2946 return r;
2947 if (r == 0) {
2948 static const char fs[] = "/run/systemd/mount-root";
2949 /* This is a child process with its own mount namespace and propagation to host turned off */
2950
2951 r = mkdir_p(fs, 0700);
2952 if (r < 0) {
2953 log_error_errno(r, "Failed to create mount point: %m");
2954 _exit(EXIT_FAILURE);
2955 }
2956
2957 if (mount_nofollow_verbose(LOG_ERR, node, fs, p->format, MS_NOATIME|MS_NODEV|MS_NOEXEC|MS_NOSUID, NULL) < 0)
2958 _exit(EXIT_FAILURE);
2959
2960 if (do_copy_files(p, fs) < 0)
2961 _exit(EXIT_FAILURE);
2962
2963 if (do_make_directories(p, fs) < 0)
2964 _exit(EXIT_FAILURE);
2965
2966 r = syncfs_path(AT_FDCWD, fs);
2967 if (r < 0) {
2968 log_error_errno(r, "Failed to synchronize written files: %m");
2969 _exit(EXIT_FAILURE);
2970 }
2971
2972 _exit(EXIT_SUCCESS);
2973 }
2974
2975 log_info("Successfully populated partition %" PRIu64 " with files.", p->partno);
2976 return 0;
2977 }
2978
2979 static int context_mkfs(Context *context) {
2980 int fd = -1, r;
2981
2982 assert(context);
2983
2984 /* Make a file system */
2985
2986 LIST_FOREACH(partitions, p, context->partitions) {
2987 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
2988 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
2989 _cleanup_free_ char *encrypted = NULL;
2990 _cleanup_close_ int encrypted_dev_fd = -1;
2991 const char *fsdev;
2992 sd_id128_t fs_uuid;
2993
2994 if (p->dropped)
2995 continue;
2996
2997 if (PARTITION_EXISTS(p)) /* Never format existing partitions */
2998 continue;
2999
3000 if (!p->format)
3001 continue;
3002
3003 assert(p->offset != UINT64_MAX);
3004 assert(p->new_size != UINT64_MAX);
3005
3006 if (fd < 0)
3007 assert_se((fd = fdisk_get_devfd(context->fdisk_context)) >= 0);
3008
3009 /* Loopback block devices are not only useful to turn regular files into block devices, but
3010 * also to cut out sections of block devices into new block devices. */
3011
3012 r = loop_device_make(fd, O_RDWR, p->offset, p->new_size, 0, &d);
3013 if (r < 0)
3014 return log_error_errno(r, "Failed to make loopback device of future partition %" PRIu64 ": %m", p->partno);
3015
3016 r = loop_device_flock(d, LOCK_EX);
3017 if (r < 0)
3018 return log_error_errno(r, "Failed to lock loopback device: %m");
3019
3020 if (p->encrypt != ENCRYPT_OFF) {
3021 r = partition_encrypt(context, p, d->node, &cd, &encrypted, &encrypted_dev_fd);
3022 if (r < 0)
3023 return log_error_errno(r, "Failed to encrypt device: %m");
3024
3025 if (flock(encrypted_dev_fd, LOCK_EX) < 0)
3026 return log_error_errno(errno, "Failed to lock LUKS device: %m");
3027
3028 fsdev = encrypted;
3029 } else
3030 fsdev = d->node;
3031
3032 log_info("Formatting future partition %" PRIu64 ".", p->partno);
3033
3034 /* Calculate the UUID for the file system as HMAC-SHA256 of the string "file-system-uuid",
3035 * keyed off the partition UUID. */
3036 r = derive_uuid(p->new_uuid, "file-system-uuid", &fs_uuid);
3037 if (r < 0)
3038 return r;
3039
3040 r = make_filesystem(fsdev, p->format, strempty(p->new_label), fs_uuid, arg_discard);
3041 if (r < 0) {
3042 encrypted_dev_fd = safe_close(encrypted_dev_fd);
3043 (void) deactivate_luks(cd, encrypted);
3044 return r;
3045 }
3046
3047 log_info("Successfully formatted future partition %" PRIu64 ".", p->partno);
3048
3049 /* The file system is now created, no need to delay udev further */
3050 if (p->encrypt != ENCRYPT_OFF)
3051 if (flock(encrypted_dev_fd, LOCK_UN) < 0)
3052 return log_error_errno(errno, "Failed to unlock LUKS device: %m");
3053
3054 r = partition_populate(p, fsdev);
3055 if (r < 0) {
3056 encrypted_dev_fd = safe_close(encrypted_dev_fd);
3057 (void) deactivate_luks(cd, encrypted);
3058 return r;
3059 }
3060
3061 /* Note that we always sync explicitly here, since mkfs.fat doesn't do that on its own, and
3062 * if we don't sync before detaching a block device the in-flight sectors possibly won't hit
3063 * the disk. */
3064
3065 if (p->encrypt != ENCRYPT_OFF) {
3066 if (fsync(encrypted_dev_fd) < 0)
3067 return log_error_errno(errno, "Failed to synchronize LUKS volume: %m");
3068 encrypted_dev_fd = safe_close(encrypted_dev_fd);
3069
3070 r = deactivate_luks(cd, encrypted);
3071 if (r < 0)
3072 return r;
3073
3074 sym_crypt_free(cd);
3075 cd = NULL;
3076 }
3077
3078 r = loop_device_sync(d);
3079 if (r < 0)
3080 return log_error_errno(r, "Failed to sync loopback device: %m");
3081 }
3082
3083 return 0;
3084 }
3085
3086 static int partition_acquire_uuid(Context *context, Partition *p, sd_id128_t *ret) {
3087 struct {
3088 sd_id128_t type_uuid;
3089 uint64_t counter;
3090 } _packed_ plaintext = {};
3091 union {
3092 uint8_t md[SHA256_DIGEST_SIZE];
3093 sd_id128_t id;
3094 } result;
3095
3096 uint64_t k = 0;
3097 int r;
3098
3099 assert(context);
3100 assert(p);
3101 assert(ret);
3102
3103 /* Calculate a good UUID for the indicated partition. We want a certain degree of reproducibility,
3104 * hence we won't generate the UUIDs randomly. Instead we use a cryptographic hash (precisely:
3105 * HMAC-SHA256) to derive them from a single seed. The seed is generally the machine ID of the
3106 * installation we are processing, but if random behaviour is desired can be random, too. We use the
3107 * seed value as key for the HMAC (since the machine ID is something we generally don't want to leak)
3108 * and the partition type as plaintext. The partition type is suffixed with a counter (only for the
3109 * second and later partition of the same type) if we have more than one partition of the same
3110 * time. Or in other words:
3111 *
3112 * With:
3113 * SEED := /etc/machine-id
3114 *
3115 * If first partition instance of type TYPE_UUID:
3116 * PARTITION_UUID := HMAC-SHA256(SEED, TYPE_UUID)
3117 *
3118 * For all later partition instances of type TYPE_UUID with INSTANCE being the LE64 encoded instance number:
3119 * PARTITION_UUID := HMAC-SHA256(SEED, TYPE_UUID || INSTANCE)
3120 */
3121
3122 LIST_FOREACH(partitions, q, context->partitions) {
3123 if (p == q)
3124 break;
3125
3126 if (!sd_id128_equal(p->type_uuid, q->type_uuid))
3127 continue;
3128
3129 k++;
3130 }
3131
3132 plaintext.type_uuid = p->type_uuid;
3133 plaintext.counter = htole64(k);
3134
3135 hmac_sha256(context->seed.bytes, sizeof(context->seed.bytes),
3136 &plaintext,
3137 k == 0 ? sizeof(sd_id128_t) : sizeof(plaintext),
3138 result.md);
3139
3140 /* Take the first half, mark it as v4 UUID */
3141 assert_cc(sizeof(result.md) == sizeof(result.id) * 2);
3142 result.id = id128_make_v4_uuid(result.id);
3143
3144 /* Ensure this partition UUID is actually unique, and there's no remaining partition from an earlier run? */
3145 LIST_FOREACH(partitions, q, context->partitions) {
3146 if (p == q)
3147 continue;
3148
3149 if (sd_id128_in_set(result.id, q->current_uuid, q->new_uuid)) {
3150 log_warning("Partition UUID calculated from seed for partition %" PRIu64 " already used, reverting to randomized UUID.", p->partno);
3151
3152 r = sd_id128_randomize(&result.id);
3153 if (r < 0)
3154 return log_error_errno(r, "Failed to generate randomized UUID: %m");
3155
3156 break;
3157 }
3158 }
3159
3160 *ret = result.id;
3161 return 0;
3162 }
3163
3164 static int partition_acquire_label(Context *context, Partition *p, char **ret) {
3165 _cleanup_free_ char *label = NULL;
3166 const char *prefix;
3167 unsigned k = 1;
3168
3169 assert(context);
3170 assert(p);
3171 assert(ret);
3172
3173 prefix = gpt_partition_type_uuid_to_string(p->type_uuid);
3174 if (!prefix)
3175 prefix = "linux";
3176
3177 for (;;) {
3178 const char *ll = label ?: prefix;
3179 bool retry = false;
3180
3181 LIST_FOREACH(partitions, q, context->partitions) {
3182 if (p == q)
3183 break;
3184
3185 if (streq_ptr(ll, q->current_label) ||
3186 streq_ptr(ll, q->new_label)) {
3187 retry = true;
3188 break;
3189 }
3190 }
3191
3192 if (!retry)
3193 break;
3194
3195 label = mfree(label);
3196 if (asprintf(&label, "%s-%u", prefix, ++k) < 0)
3197 return log_oom();
3198 }
3199
3200 if (!label) {
3201 label = strdup(prefix);
3202 if (!label)
3203 return log_oom();
3204 }
3205
3206 *ret = TAKE_PTR(label);
3207 return 0;
3208 }
3209
3210 static int context_acquire_partition_uuids_and_labels(Context *context) {
3211 int r;
3212
3213 assert(context);
3214
3215 LIST_FOREACH(partitions, p, context->partitions) {
3216 /* Never touch foreign partitions */
3217 if (PARTITION_IS_FOREIGN(p)) {
3218 p->new_uuid = p->current_uuid;
3219
3220 if (p->current_label) {
3221 r = free_and_strdup_warn(&p->new_label, strempty(p->current_label));
3222 if (r < 0)
3223 return r;
3224 }
3225
3226 continue;
3227 }
3228
3229 if (!sd_id128_is_null(p->current_uuid))
3230 p->new_uuid = p->current_uuid; /* Never change initialized UUIDs */
3231 else if (sd_id128_is_null(p->new_uuid)) {
3232 /* Not explicitly set by user! */
3233 r = partition_acquire_uuid(context, p, &p->new_uuid);
3234 if (r < 0)
3235 return r;
3236 }
3237
3238 if (!isempty(p->current_label)) {
3239 /* never change initialized labels */
3240 r = free_and_strdup_warn(&p->new_label, p->current_label);
3241 if (r < 0)
3242 return r;
3243 } else if (!p->new_label) {
3244 /* Not explicitly set by user! */
3245
3246 r = partition_acquire_label(context, p, &p->new_label);
3247 if (r < 0)
3248 return r;
3249 }
3250 }
3251
3252 return 0;
3253 }
3254
3255 static int set_gpt_flags(struct fdisk_partition *q, uint64_t flags) {
3256 _cleanup_free_ char *a = NULL;
3257
3258 for (unsigned i = 0; i < sizeof(flags) * 8; i++) {
3259 uint64_t bit = UINT64_C(1) << i;
3260 char buf[DECIMAL_STR_MAX(unsigned)+1];
3261
3262 if (!FLAGS_SET(flags, bit))
3263 continue;
3264
3265 xsprintf(buf, "%u", i);
3266 if (!strextend_with_separator(&a, ",", buf))
3267 return -ENOMEM;
3268 }
3269
3270 return fdisk_partition_set_attrs(q, a);
3271 }
3272
3273 static uint64_t partition_merge_flags(Partition *p) {
3274 uint64_t f;
3275
3276 assert(p);
3277
3278 f = p->gpt_flags;
3279
3280 if (p->no_auto >= 0) {
3281 if (gpt_partition_type_knows_no_auto(p->type_uuid))
3282 SET_FLAG(f, GPT_FLAG_NO_AUTO, p->no_auto);
3283 else {
3284 char buffer[SD_ID128_UUID_STRING_MAX];
3285 log_warning("Configured NoAuto=%s for partition type '%s' that doesn't support it, ignoring.",
3286 yes_no(p->no_auto),
3287 gpt_partition_type_uuid_to_string_harder(p->type_uuid, buffer));
3288 }
3289 }
3290
3291 if (p->read_only >= 0) {
3292 if (gpt_partition_type_knows_read_only(p->type_uuid))
3293 SET_FLAG(f, GPT_FLAG_READ_ONLY, p->read_only);
3294 else {
3295 char buffer[SD_ID128_UUID_STRING_MAX];
3296 log_warning("Configured ReadOnly=%s for partition type '%s' that doesn't support it, ignoring.",
3297 yes_no(p->read_only),
3298 gpt_partition_type_uuid_to_string_harder(p->type_uuid, buffer));
3299 }
3300 }
3301
3302 if (p->growfs >= 0) {
3303 if (gpt_partition_type_knows_growfs(p->type_uuid))
3304 SET_FLAG(f, GPT_FLAG_GROWFS, p->growfs);
3305 else {
3306 char buffer[SD_ID128_UUID_STRING_MAX];
3307 log_warning("Configured GrowFileSystem=%s for partition type '%s' that doesn't support it, ignoring.",
3308 yes_no(p->growfs),
3309 gpt_partition_type_uuid_to_string_harder(p->type_uuid, buffer));
3310 }
3311 }
3312
3313 return f;
3314 }
3315
3316 static int context_mangle_partitions(Context *context) {
3317 int r;
3318
3319 assert(context);
3320
3321 LIST_FOREACH(partitions, p, context->partitions) {
3322 if (p->dropped)
3323 continue;
3324
3325 assert(p->new_size != UINT64_MAX);
3326 assert(p->offset != UINT64_MAX);
3327 assert(p->partno != UINT64_MAX);
3328
3329 if (PARTITION_EXISTS(p)) {
3330 bool changed = false;
3331
3332 assert(p->current_partition);
3333
3334 if (p->new_size != p->current_size) {
3335 assert(p->new_size >= p->current_size);
3336 assert(p->new_size % context->sector_size == 0);
3337
3338 r = fdisk_partition_size_explicit(p->current_partition, true);
3339 if (r < 0)
3340 return log_error_errno(r, "Failed to enable explicit sizing: %m");
3341
3342 r = fdisk_partition_set_size(p->current_partition, p->new_size / context->sector_size);
3343 if (r < 0)
3344 return log_error_errno(r, "Failed to grow partition: %m");
3345
3346 log_info("Growing existing partition %" PRIu64 ".", p->partno);
3347 changed = true;
3348 }
3349
3350 if (!sd_id128_equal(p->new_uuid, p->current_uuid)) {
3351 assert(!sd_id128_is_null(p->new_uuid));
3352
3353 r = fdisk_partition_set_uuid(p->current_partition, SD_ID128_TO_UUID_STRING(p->new_uuid));
3354 if (r < 0)
3355 return log_error_errno(r, "Failed to set partition UUID: %m");
3356
3357 log_info("Initializing UUID of existing partition %" PRIu64 ".", p->partno);
3358 changed = true;
3359 }
3360
3361 if (!streq_ptr(p->new_label, p->current_label)) {
3362 r = fdisk_partition_set_name(p->current_partition, strempty(p->new_label));
3363 if (r < 0)
3364 return log_error_errno(r, "Failed to set partition label: %m");
3365
3366 log_info("Setting partition label of existing partition %" PRIu64 ".", p->partno);
3367 changed = true;
3368 }
3369
3370 if (changed) {
3371 assert(!PARTITION_IS_FOREIGN(p)); /* never touch foreign partitions */
3372
3373 r = fdisk_set_partition(context->fdisk_context, p->partno, p->current_partition);
3374 if (r < 0)
3375 return log_error_errno(r, "Failed to update partition: %m");
3376 }
3377 } else {
3378 _cleanup_(fdisk_unref_partitionp) struct fdisk_partition *q = NULL;
3379 _cleanup_(fdisk_unref_parttypep) struct fdisk_parttype *t = NULL;
3380
3381 assert(!p->new_partition);
3382 assert(p->offset % context->sector_size == 0);
3383 assert(p->new_size % context->sector_size == 0);
3384 assert(!sd_id128_is_null(p->new_uuid));
3385 assert(p->new_label);
3386
3387 t = fdisk_new_parttype();
3388 if (!t)
3389 return log_oom();
3390
3391 r = fdisk_parttype_set_typestr(t, SD_ID128_TO_UUID_STRING(p->type_uuid));
3392 if (r < 0)
3393 return log_error_errno(r, "Failed to initialize partition type: %m");
3394
3395 q = fdisk_new_partition();
3396 if (!q)
3397 return log_oom();
3398
3399 r = fdisk_partition_set_type(q, t);
3400 if (r < 0)
3401 return log_error_errno(r, "Failed to set partition type: %m");
3402
3403 r = fdisk_partition_size_explicit(q, true);
3404 if (r < 0)
3405 return log_error_errno(r, "Failed to enable explicit sizing: %m");
3406
3407 r = fdisk_partition_set_start(q, p->offset / context->sector_size);
3408 if (r < 0)
3409 return log_error_errno(r, "Failed to position partition: %m");
3410
3411 r = fdisk_partition_set_size(q, p->new_size / context->sector_size);
3412 if (r < 0)
3413 return log_error_errno(r, "Failed to grow partition: %m");
3414
3415 r = fdisk_partition_set_partno(q, p->partno);
3416 if (r < 0)
3417 return log_error_errno(r, "Failed to set partition number: %m");
3418
3419 r = fdisk_partition_set_uuid(q, SD_ID128_TO_UUID_STRING(p->new_uuid));
3420 if (r < 0)
3421 return log_error_errno(r, "Failed to set partition UUID: %m");
3422
3423 r = fdisk_partition_set_name(q, strempty(p->new_label));
3424 if (r < 0)
3425 return log_error_errno(r, "Failed to set partition label: %m");
3426
3427 /* Merge the no auto + read only + growfs setting with the literal flags, and set them for the partition */
3428 r = set_gpt_flags(q, partition_merge_flags(p));
3429 if (r < 0)
3430 return log_error_errno(r, "Failed to set GPT partition flags: %m");
3431
3432 log_info("Adding new partition %" PRIu64 " to partition table.", p->partno);
3433
3434 r = fdisk_add_partition(context->fdisk_context, q, NULL);
3435 if (r < 0)
3436 return log_error_errno(r, "Failed to add partition: %m");
3437
3438 assert(!p->new_partition);
3439 p->new_partition = TAKE_PTR(q);
3440 }
3441 }
3442
3443 return 0;
3444 }
3445
3446 static int context_write_partition_table(
3447 Context *context,
3448 const char *node,
3449 bool from_scratch) {
3450
3451 _cleanup_(fdisk_unref_tablep) struct fdisk_table *original_table = NULL;
3452 int capable, r;
3453
3454 assert(context);
3455
3456 if (arg_pretty > 0 ||
3457 (arg_pretty < 0 && isatty(STDOUT_FILENO) > 0) ||
3458 !FLAGS_SET(arg_json_format_flags, JSON_FORMAT_OFF)) {
3459
3460 (void) context_dump_partitions(context, node);
3461
3462 putc('\n', stdout);
3463
3464 if (arg_json_format_flags & JSON_FORMAT_OFF)
3465 (void) context_dump_partition_bar(context, node);
3466 putc('\n', stdout);
3467 fflush(stdout);
3468 }
3469
3470 if (!from_scratch && !context_changed(context)) {
3471 log_info("No changes.");
3472 return 0;
3473 }
3474
3475 if (arg_dry_run) {
3476 log_notice("Refusing to repartition, please re-run with --dry-run=no.");
3477 return 0;
3478 }
3479
3480 log_info("Applying changes.");
3481
3482 if (from_scratch) {
3483 r = context_wipe_range(context, 0, context->total);
3484 if (r < 0)
3485 return r;
3486
3487 log_info("Wiped block device.");
3488
3489 r = context_discard_range(context, 0, context->total);
3490 if (r == -EOPNOTSUPP)
3491 log_info("Storage does not support discard, not discarding entire block device data.");
3492 else if (r < 0)
3493 return log_error_errno(r, "Failed to discard entire block device: %m");
3494 else if (r > 0)
3495 log_info("Discarded entire block device.");
3496 }
3497
3498 r = fdisk_get_partitions(context->fdisk_context, &original_table);
3499 if (r < 0)
3500 return log_error_errno(r, "Failed to acquire partition table: %m");
3501
3502 /* Wipe fs signatures and discard sectors where the new partitions are going to be placed and in the
3503 * gaps between partitions, just to be sure. */
3504 r = context_wipe_and_discard(context, from_scratch);
3505 if (r < 0)
3506 return r;
3507
3508 r = context_copy_blocks(context);
3509 if (r < 0)
3510 return r;
3511
3512 r = context_mkfs(context);
3513 if (r < 0)
3514 return r;
3515
3516 r = context_mangle_partitions(context);
3517 if (r < 0)
3518 return r;
3519
3520 log_info("Writing new partition table.");
3521
3522 r = fdisk_write_disklabel(context->fdisk_context);
3523 if (r < 0)
3524 return log_error_errno(r, "Failed to write partition table: %m");
3525
3526 capable = blockdev_partscan_enabled(fdisk_get_devfd(context->fdisk_context));
3527 if (capable == -ENOTBLK)
3528 log_debug("Not telling kernel to reread partition table, since we are not operating on a block device.");
3529 else if (capable < 0)
3530 return log_error_errno(capable, "Failed to check if block device supports partition scanning: %m");
3531 else if (capable > 0) {
3532 log_info("Telling kernel to reread partition table.");
3533
3534 if (from_scratch)
3535 r = fdisk_reread_partition_table(context->fdisk_context);
3536 else
3537 r = fdisk_reread_changes(context->fdisk_context, original_table);
3538 if (r < 0)
3539 return log_error_errno(r, "Failed to reread partition table: %m");
3540 } else
3541 log_notice("Not telling kernel to reread partition table, because selected image does not support kernel partition block devices.");
3542
3543 log_info("All done.");
3544
3545 return 0;
3546 }
3547
3548 static int context_read_seed(Context *context, const char *root) {
3549 int r;
3550
3551 assert(context);
3552
3553 if (!sd_id128_is_null(context->seed))
3554 return 0;
3555
3556 if (!arg_randomize) {
3557 _cleanup_close_ int fd = -1;
3558
3559 fd = chase_symlinks_and_open("/etc/machine-id", root, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC, NULL);
3560 if (fd == -ENOENT)
3561 log_info("No machine ID set, using randomized partition UUIDs.");
3562 else if (fd < 0)
3563 return log_error_errno(fd, "Failed to determine machine ID of image: %m");
3564 else {
3565 r = id128_read_fd(fd, ID128_PLAIN_OR_UNINIT, &context->seed);
3566 if (r == -ENOMEDIUM)
3567 log_info("No machine ID set, using randomized partition UUIDs.");
3568 else if (r < 0)
3569 return log_error_errno(r, "Failed to parse machine ID of image: %m");
3570
3571 return 0;
3572 }
3573 }
3574
3575 r = sd_id128_randomize(&context->seed);
3576 if (r < 0)
3577 return log_error_errno(r, "Failed to generate randomized seed: %m");
3578
3579 return 0;
3580 }
3581
3582 static int context_factory_reset(Context *context, bool from_scratch) {
3583 size_t n = 0;
3584 int r;
3585
3586 assert(context);
3587
3588 if (arg_factory_reset <= 0)
3589 return 0;
3590
3591 if (from_scratch) /* Nothing to reset if we start from scratch */
3592 return 0;
3593
3594 if (arg_dry_run) {
3595 log_notice("Refusing to factory reset, please re-run with --dry-run=no.");
3596 return 0;
3597 }
3598
3599 log_info("Applying factory reset.");
3600
3601 LIST_FOREACH(partitions, p, context->partitions) {
3602
3603 if (!p->factory_reset || !PARTITION_EXISTS(p))
3604 continue;
3605
3606 assert(p->partno != UINT64_MAX);
3607
3608 log_info("Removing partition %" PRIu64 " for factory reset.", p->partno);
3609
3610 r = fdisk_delete_partition(context->fdisk_context, p->partno);
3611 if (r < 0)
3612 return log_error_errno(r, "Failed to remove partition %" PRIu64 ": %m", p->partno);
3613
3614 n++;
3615 }
3616
3617 if (n == 0) {
3618 log_info("Factory reset requested, but no partitions to delete found.");
3619 return 0;
3620 }
3621
3622 r = fdisk_write_disklabel(context->fdisk_context);
3623 if (r < 0)
3624 return log_error_errno(r, "Failed to write disk label: %m");
3625
3626 log_info("Successfully deleted %zu partitions.", n);
3627 return 1;
3628 }
3629
3630 static int context_can_factory_reset(Context *context) {
3631 assert(context);
3632
3633 LIST_FOREACH(partitions, p, context->partitions)
3634 if (p->factory_reset && PARTITION_EXISTS(p))
3635 return true;
3636
3637 return false;
3638 }
3639
3640 static int resolve_copy_blocks_auto_candidate(
3641 dev_t partition_devno,
3642 sd_id128_t partition_type_uuid,
3643 dev_t restrict_devno,
3644 sd_id128_t *ret_uuid) {
3645
3646 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
3647 _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
3648 _cleanup_close_ int fd = -1;
3649 const char *pttype, *t, *p;
3650 sd_id128_t pt_parsed, u;
3651 blkid_partition pp;
3652 dev_t whole_devno;
3653 blkid_partlist pl;
3654 int r;
3655
3656 /* Checks if the specified partition has the specified GPT type UUID, and is located on the specified
3657 * 'restrict_devno' device. The type check is particularly relevant if we have Verity volume which is
3658 * backed by two separate partitions: the data and the hash partitions, and we need to find the right
3659 * one of the two. */
3660
3661 r = block_get_whole_disk(partition_devno, &whole_devno);
3662 if (r < 0)
3663 return log_error_errno(
3664 r,
3665 "Unable to determine containing block device of partition %u:%u: %m",
3666 major(partition_devno), minor(partition_devno));
3667
3668 if (restrict_devno != (dev_t) -1 &&
3669 restrict_devno != whole_devno)
3670 return log_error_errno(
3671 SYNTHETIC_ERRNO(EPERM),
3672 "Partition %u:%u is located outside of block device %u:%u, refusing.",
3673 major(partition_devno), minor(partition_devno),
3674 major(restrict_devno), minor(restrict_devno));
3675
3676 r = sd_device_new_from_devnum(&dev, 'b', whole_devno);
3677 if (r < 0)
3678 return log_error_errno(r, "Failed to create sd-device for block device %u:%u: %m",
3679 major(whole_devno), minor(whole_devno));
3680
3681 r = sd_device_get_devname(dev, &p);
3682 if (r < 0)
3683 return log_error_errno(r, "Failed to get name of block device %u:%u: %m",
3684 major(whole_devno), minor(whole_devno));
3685
3686 fd = sd_device_open(dev, O_RDONLY|O_CLOEXEC|O_NONBLOCK);
3687 if (fd < 0)
3688 return log_error_errno(fd, "Failed to open block device %s: %m", p);
3689
3690 b = blkid_new_probe();
3691 if (!b)
3692 return log_oom();
3693
3694 errno = 0;
3695 r = blkid_probe_set_device(b, fd, 0, 0);
3696 if (r != 0)
3697 return log_error_errno(errno_or_else(ENOMEM), "Failed to open block device '%s': %m", p);
3698
3699 (void) blkid_probe_enable_partitions(b, 1);
3700 (void) blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
3701
3702 errno = 0;
3703 r = blkid_do_safeprobe(b);
3704 if (IN_SET(r, -2, 1)) { /* nothing found or ambiguous result */
3705 log_debug("Didn't find partition table on block device '%s'.", p);
3706 return false;
3707 }
3708 if (r != 0)
3709 return log_error_errno(errno_or_else(EIO), "Unable to probe for partition table of '%s': %m", p);
3710
3711 (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
3712 if (!streq_ptr(pttype, "gpt")) {
3713 log_debug("Didn't find a GPT partition table on '%s'.", p);
3714 return false;
3715 }
3716
3717 errno = 0;
3718 pl = blkid_probe_get_partitions(b);
3719 if (!pl)
3720 return log_error_errno(errno_or_else(EIO), "Unable read partition table of '%s': %m", p);
3721 errno = 0;
3722
3723 pp = blkid_partlist_devno_to_partition(pl, partition_devno);
3724 if (!pp) {
3725 log_debug("Partition %u:%u has no matching partition table entry on '%s'.",
3726 major(partition_devno), minor(partition_devno), p);
3727 return false;
3728 }
3729
3730 t = blkid_partition_get_type_string(pp);
3731 if (isempty(t)) {
3732 log_debug("Partition %u:%u has no type on '%s'.",
3733 major(partition_devno), minor(partition_devno), p);
3734 return false;
3735 }
3736
3737 r = sd_id128_from_string(t, &pt_parsed);
3738 if (r < 0) {
3739 log_debug_errno(r, "Failed to parse partition type \"%s\": %m", t);
3740 return false;
3741 }
3742
3743 if (!sd_id128_equal(pt_parsed, partition_type_uuid)) {
3744 log_debug("Partition %u:%u has non-matching partition type " SD_ID128_FORMAT_STR " (needed: " SD_ID128_FORMAT_STR "), ignoring.",
3745 major(partition_devno), minor(partition_devno),
3746 SD_ID128_FORMAT_VAL(pt_parsed), SD_ID128_FORMAT_VAL(partition_type_uuid));
3747 return false;
3748 }
3749
3750 t = blkid_partition_get_uuid(pp);
3751 if (isempty(t)) {
3752 log_debug("Partition %u:%u has no UUID.",
3753 major(partition_devno), minor(partition_devno));
3754 return false;
3755 }
3756
3757 r = sd_id128_from_string(t, &u);
3758 if (r < 0) {
3759 log_debug_errno(r, "Failed to parse partition UUID \"%s\": %m", t);
3760 return false;
3761 }
3762
3763 log_debug("Automatically found partition %u:%u of right type " SD_ID128_FORMAT_STR ".",
3764 major(partition_devno), minor(partition_devno),
3765 SD_ID128_FORMAT_VAL(pt_parsed));
3766
3767 if (ret_uuid)
3768 *ret_uuid = u;
3769
3770 return true;
3771 }
3772
3773 static int find_backing_devno(
3774 const char *path,
3775 const char *root,
3776 dev_t *ret) {
3777
3778 _cleanup_free_ char *resolved = NULL;
3779 int r;
3780
3781 assert(path);
3782
3783 r = chase_symlinks(path, root, CHASE_PREFIX_ROOT, &resolved, NULL);
3784 if (r < 0)
3785 return r;
3786
3787 r = path_is_mount_point(resolved, NULL, 0);
3788 if (r < 0)
3789 return r;
3790 if (r == 0) /* Not a mount point, then it's not a partition of its own, let's not automatically use it. */
3791 return -ENOENT;
3792
3793 r = get_block_device(resolved, ret);
3794 if (r < 0)
3795 return r;
3796 if (r == 0) /* Not backed by physical file system, we can't use this */
3797 return -ENOENT;
3798
3799 return 0;
3800 }
3801
3802 static int resolve_copy_blocks_auto(
3803 sd_id128_t type_uuid,
3804 const char *root,
3805 dev_t restrict_devno,
3806 dev_t *ret_devno,
3807 sd_id128_t *ret_uuid) {
3808
3809 const char *try1 = NULL, *try2 = NULL;
3810 char p[SYS_BLOCK_PATH_MAX("/slaves")];
3811 _cleanup_(closedirp) DIR *d = NULL;
3812 sd_id128_t found_uuid = SD_ID128_NULL;
3813 dev_t devno, found = 0;
3814 int r;
3815
3816 /* Enforce some security restrictions: CopyBlocks=auto should not be an avenue to get outside of the
3817 * --root=/--image= confinement. Specifically, refuse CopyBlocks= in combination with --root= at all,
3818 * and restrict block device references in the --image= case to loopback block device we set up.
3819 *
3820 * restrict_devno contain the dev_t of the loop back device we operate on in case of --image=, and
3821 * thus declares which device (and its partition subdevices) we shall limit access to. If
3822 * restrict_devno is zero no device probing access shall be allowed at all (used for --root=) and if
3823 * it is (dev_t) -1 then free access shall be allowed (if neither switch is used). */
3824
3825 if (restrict_devno == 0)
3826 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
3827 "Automatic discovery of backing block devices not permitted in --root= mode, refusing.");
3828
3829 /* Handles CopyBlocks=auto, and finds the right source partition to copy from. We look for matching
3830 * partitions in the host, using the appropriate directory as key and ensuring that the partition
3831 * type matches. */
3832
3833 if (gpt_partition_type_is_root(type_uuid))
3834 try1 = "/";
3835 else if (gpt_partition_type_is_usr(type_uuid))
3836 try1 = "/usr/";
3837 else if (gpt_partition_type_is_root_verity(type_uuid))
3838 try1 = "/";
3839 else if (gpt_partition_type_is_usr_verity(type_uuid))
3840 try1 = "/usr/";
3841 else if (sd_id128_equal(type_uuid, GPT_ESP)) {
3842 try1 = "/efi/";
3843 try2 = "/boot/";
3844 } else if (sd_id128_equal(type_uuid, GPT_XBOOTLDR))
3845 try1 = "/boot/";
3846 else
3847 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
3848 "Partition type " SD_ID128_FORMAT_STR " not supported from automatic source block device discovery.",
3849 SD_ID128_FORMAT_VAL(type_uuid));
3850
3851 r = find_backing_devno(try1, root, &devno);
3852 if (r == -ENOENT && try2)
3853 r = find_backing_devno(try2, root, &devno);
3854 if (r < 0)
3855 return log_error_errno(r, "Failed to resolve automatic CopyBlocks= path for partition type " SD_ID128_FORMAT_STR ", sorry: %m",
3856 SD_ID128_FORMAT_VAL(type_uuid));
3857
3858 xsprintf_sys_block_path(p, "/slaves", devno);
3859 d = opendir(p);
3860 if (d) {
3861 struct dirent *de;
3862
3863 for (;;) {
3864 _cleanup_free_ char *q = NULL, *t = NULL;
3865 sd_id128_t u;
3866 dev_t sl;
3867
3868 errno = 0;
3869 de = readdir_no_dot(d);
3870 if (!de) {
3871 if (errno != 0)
3872 return log_error_errno(errno, "Failed to read directory '%s': %m", p);
3873
3874 break;
3875 }
3876
3877 if (!IN_SET(de->d_type, DT_LNK, DT_UNKNOWN))
3878 continue;
3879
3880 q = path_join(p, de->d_name, "/dev");
3881 if (!q)
3882 return log_oom();
3883
3884 r = read_one_line_file(q, &t);
3885 if (r < 0)
3886 return log_error_errno(r, "Failed to read %s: %m", q);
3887
3888 r = parse_devnum(t, &sl);
3889 if (r < 0) {
3890 log_debug_errno(r, "Failed to parse %s, ignoring: %m", q);
3891 continue;
3892 }
3893 if (major(sl) == 0) {
3894 log_debug_errno(r, "Device backing %s is special, ignoring: %m", q);
3895 continue;
3896 }
3897
3898 r = resolve_copy_blocks_auto_candidate(sl, type_uuid, restrict_devno, &u);
3899 if (r < 0)
3900 return r;
3901 if (r > 0) {
3902 /* We found a matching one! */
3903 if (found != 0)
3904 return log_error_errno(SYNTHETIC_ERRNO(ENOTUNIQ),
3905 "Multiple matching partitions found, refusing.");
3906
3907 found = sl;
3908 found_uuid = u;
3909 }
3910 }
3911 } else if (errno != ENOENT)
3912 return log_error_errno(errno, "Failed open %s: %m", p);
3913 else {
3914 r = resolve_copy_blocks_auto_candidate(devno, type_uuid, restrict_devno, &found_uuid);
3915 if (r < 0)
3916 return r;
3917 if (r > 0)
3918 found = devno;
3919 }
3920
3921 if (found == 0)
3922 return log_error_errno(SYNTHETIC_ERRNO(ENXIO),
3923 "Unable to automatically discover suitable partition to copy blocks from.");
3924
3925 if (ret_devno)
3926 *ret_devno = found;
3927
3928 if (ret_uuid)
3929 *ret_uuid = found_uuid;
3930
3931 return 0;
3932 }
3933
3934 static int context_open_copy_block_paths(
3935 Context *context,
3936 const char *root,
3937 dev_t restrict_devno) {
3938
3939 int r;
3940
3941 assert(context);
3942
3943 LIST_FOREACH(partitions, p, context->partitions) {
3944 _cleanup_close_ int source_fd = -1;
3945 _cleanup_free_ char *opened = NULL;
3946 sd_id128_t uuid = SD_ID128_NULL;
3947 uint64_t size;
3948 struct stat st;
3949
3950 assert(p->copy_blocks_fd < 0);
3951 assert(p->copy_blocks_size == UINT64_MAX);
3952
3953 if (PARTITION_EXISTS(p)) /* Never copy over partitions that already exist! */
3954 continue;
3955
3956 if (p->copy_blocks_path) {
3957
3958 source_fd = chase_symlinks_and_open(p->copy_blocks_path, root, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NONBLOCK, &opened);
3959 if (source_fd < 0)
3960 return log_error_errno(source_fd, "Failed to open '%s': %m", p->copy_blocks_path);
3961
3962 if (fstat(source_fd, &st) < 0)
3963 return log_error_errno(errno, "Failed to stat block copy file '%s': %m", opened);
3964
3965 if (!S_ISREG(st.st_mode) && restrict_devno != (dev_t) -1)
3966 return log_error_errno(SYNTHETIC_ERRNO(EPERM),
3967 "Copying from block device node is not permitted in --image=/--root= mode, refusing.");
3968
3969 } else if (p->copy_blocks_auto) {
3970 _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
3971 const char *devname;
3972 dev_t devno;
3973
3974 r = resolve_copy_blocks_auto(p->type_uuid, root, restrict_devno, &devno, &uuid);
3975 if (r < 0)
3976 return r;
3977
3978 r = sd_device_new_from_devnum(&dev, 'b', devno);
3979 if (r < 0)
3980 return log_error_errno(r, "Failed to create sd-device object for device %u:%u: %m", major(devno), minor(devno));
3981
3982 r = sd_device_get_devname(dev, &devname);
3983 if (r < 0)
3984 return log_error_errno(r, "Failed to get device name of %u:%u: %m", major(devno), minor(devno));
3985
3986 opened = strdup(devname);
3987 if (!opened)
3988 return log_oom();
3989
3990 source_fd = sd_device_open(dev, O_RDONLY|O_CLOEXEC|O_NONBLOCK);
3991 if (source_fd < 0)
3992 return log_error_errno(source_fd, "Failed to open automatically determined source block copy device '%s': %m", opened);
3993
3994 if (fstat(source_fd, &st) < 0)
3995 return log_error_errno(errno, "Failed to stat block copy file '%s': %m", opened);
3996 } else
3997 continue;
3998
3999 if (S_ISDIR(st.st_mode)) {
4000 _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
4001 const char *bdev;
4002
4003 /* If the file is a directory, automatically find the backing block device */
4004
4005 if (major(st.st_dev) != 0)
4006 r = sd_device_new_from_devnum(&dev, 'b', st.st_dev);
4007 else {
4008 dev_t devt;
4009
4010 /* Special support for btrfs */
4011
4012 r = btrfs_get_block_device_fd(source_fd, &devt);
4013 if (r == -EUCLEAN)
4014 return btrfs_log_dev_root(LOG_ERR, r, opened);
4015 if (r < 0)
4016 return log_error_errno(r, "Unable to determine backing block device of '%s': %m", opened);
4017
4018 r = sd_device_new_from_devnum(&dev, 'b', devt);
4019 }
4020 if (r < 0)
4021 return log_error_errno(r, "Failed to create sd-device object for block device backing '%s': %m", opened);
4022
4023 r = sd_device_get_devpath(dev, &bdev);
4024 if (r < 0)
4025 return log_error_errno(r, "Failed to get device name for block device backing '%s': %m", opened);
4026
4027 safe_close(source_fd);
4028
4029 source_fd = sd_device_open(dev, O_RDONLY|O_CLOEXEC|O_NONBLOCK);
4030 if (source_fd < 0)
4031 return log_error_errno(source_fd, "Failed to open block device '%s': %m", bdev);
4032
4033 if (fstat(source_fd, &st) < 0)
4034 return log_error_errno(errno, "Failed to stat block device '%s': %m", bdev);
4035 }
4036
4037 if (S_ISREG(st.st_mode))
4038 size = st.st_size;
4039 else if (S_ISBLK(st.st_mode)) {
4040 if (ioctl(source_fd, BLKGETSIZE64, &size) != 0)
4041 return log_error_errno(errno, "Failed to determine size of block device to copy from: %m");
4042 } else
4043 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Specified path to copy blocks from '%s' is not a regular file, block device or directory, refusing: %m", opened);
4044
4045 if (size <= 0)
4046 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "File to copy bytes from '%s' has zero size, refusing.", opened);
4047 if (size % 512 != 0)
4048 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "File to copy bytes from '%s' has size that is not multiple of 512, refusing.", opened);
4049
4050 p->copy_blocks_fd = TAKE_FD(source_fd);
4051 p->copy_blocks_size = size;
4052
4053 free_and_replace(p->copy_blocks_path, opened);
4054
4055 /* When copying from an existing partition copy that partitions UUID if none is configured explicitly */
4056 if (sd_id128_is_null(p->new_uuid) && !sd_id128_is_null(uuid))
4057 p->new_uuid = uuid;
4058 }
4059
4060 return 0;
4061 }
4062
4063 static int help(void) {
4064 _cleanup_free_ char *link = NULL;
4065 int r;
4066
4067 r = terminal_urlify_man("systemd-repart", "1", &link);
4068 if (r < 0)
4069 return log_oom();
4070
4071 printf("%s [OPTIONS...] [DEVICE]\n"
4072 "\n%sGrow and add partitions to partition table.%s\n\n"
4073 " -h --help Show this help\n"
4074 " --version Show package version\n"
4075 " --no-pager Do not pipe output into a pager\n"
4076 " --no-legend Do not show the headers and footers\n"
4077 " --dry-run=BOOL Whether to run dry-run operation\n"
4078 " --empty=MODE One of refuse, allow, require, force, create; controls\n"
4079 " how to handle empty disks lacking partition tables\n"
4080 " --discard=BOOL Whether to discard backing blocks for new partitions\n"
4081 " --pretty=BOOL Whether to show pretty summary before doing changes\n"
4082 " --factory-reset=BOOL Whether to remove data partitions before recreating\n"
4083 " them\n"
4084 " --can-factory-reset Test whether factory reset is defined\n"
4085 " --root=PATH Operate relative to root path\n"
4086 " --image=PATH Operate relative to image file\n"
4087 " --definitions=DIR Find partition definitions in specified directory\n"
4088 " --key-file=PATH Key to use when encrypting partitions\n"
4089 " --tpm2-device=PATH Path to TPM2 device node to use\n"
4090 " --tpm2-pcrs=PCR1+PCR2+PCR3+…\n"
4091 " TPM2 PCR indexes to use for TPM2 enrollment\n"
4092 " --seed=UUID 128bit seed UUID to derive all UUIDs from\n"
4093 " --size=BYTES Grow loopback file to specified size\n"
4094 " --json=pretty|short|off\n"
4095 " Generate JSON output\n"
4096 "\nSee the %s for details.\n",
4097 program_invocation_short_name,
4098 ansi_highlight(),
4099 ansi_normal(),
4100 link);
4101
4102 return 0;
4103 }
4104
4105 static int parse_argv(int argc, char *argv[]) {
4106
4107 enum {
4108 ARG_VERSION = 0x100,
4109 ARG_NO_PAGER,
4110 ARG_NO_LEGEND,
4111 ARG_DRY_RUN,
4112 ARG_EMPTY,
4113 ARG_DISCARD,
4114 ARG_FACTORY_RESET,
4115 ARG_CAN_FACTORY_RESET,
4116 ARG_ROOT,
4117 ARG_IMAGE,
4118 ARG_SEED,
4119 ARG_PRETTY,
4120 ARG_DEFINITIONS,
4121 ARG_SIZE,
4122 ARG_JSON,
4123 ARG_KEY_FILE,
4124 ARG_TPM2_DEVICE,
4125 ARG_TPM2_PCRS,
4126 };
4127
4128 static const struct option options[] = {
4129 { "help", no_argument, NULL, 'h' },
4130 { "version", no_argument, NULL, ARG_VERSION },
4131 { "no-pager", no_argument, NULL, ARG_NO_PAGER },
4132 { "no-legend", no_argument, NULL, ARG_NO_LEGEND },
4133 { "dry-run", required_argument, NULL, ARG_DRY_RUN },
4134 { "empty", required_argument, NULL, ARG_EMPTY },
4135 { "discard", required_argument, NULL, ARG_DISCARD },
4136 { "factory-reset", required_argument, NULL, ARG_FACTORY_RESET },
4137 { "can-factory-reset", no_argument, NULL, ARG_CAN_FACTORY_RESET },
4138 { "root", required_argument, NULL, ARG_ROOT },
4139 { "image", required_argument, NULL, ARG_IMAGE },
4140 { "seed", required_argument, NULL, ARG_SEED },
4141 { "pretty", required_argument, NULL, ARG_PRETTY },
4142 { "definitions", required_argument, NULL, ARG_DEFINITIONS },
4143 { "size", required_argument, NULL, ARG_SIZE },
4144 { "json", required_argument, NULL, ARG_JSON },
4145 { "key-file", required_argument, NULL, ARG_KEY_FILE },
4146 { "tpm2-device", required_argument, NULL, ARG_TPM2_DEVICE },
4147 { "tpm2-pcrs", required_argument, NULL, ARG_TPM2_PCRS },
4148 {}
4149 };
4150
4151 int c, r, dry_run = -1;
4152
4153 assert(argc >= 0);
4154 assert(argv);
4155
4156 while ((c = getopt_long(argc, argv, "h", options, NULL)) >= 0)
4157
4158 switch (c) {
4159
4160 case 'h':
4161 return help();
4162
4163 case ARG_VERSION:
4164 return version();
4165
4166 case ARG_NO_PAGER:
4167 arg_pager_flags |= PAGER_DISABLE;
4168 break;
4169
4170 case ARG_NO_LEGEND:
4171 arg_legend = false;
4172 break;
4173
4174 case ARG_DRY_RUN:
4175 r = parse_boolean_argument("--dry-run=", optarg, &arg_dry_run);
4176 if (r < 0)
4177 return r;
4178 break;
4179
4180 case ARG_EMPTY:
4181 if (isempty(optarg) || streq(optarg, "refuse"))
4182 arg_empty = EMPTY_REFUSE;
4183 else if (streq(optarg, "allow"))
4184 arg_empty = EMPTY_ALLOW;
4185 else if (streq(optarg, "require"))
4186 arg_empty = EMPTY_REQUIRE;
4187 else if (streq(optarg, "force"))
4188 arg_empty = EMPTY_FORCE;
4189 else if (streq(optarg, "create")) {
4190 arg_empty = EMPTY_CREATE;
4191
4192 if (dry_run < 0)
4193 dry_run = false; /* Imply --dry-run=no if we create the loopback file
4194 * anew. After all we cannot really break anyone's
4195 * partition tables that way. */
4196 } else
4197 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4198 "Failed to parse --empty= parameter: %s", optarg);
4199 break;
4200
4201 case ARG_DISCARD:
4202 r = parse_boolean_argument("--discard=", optarg, &arg_discard);
4203 if (r < 0)
4204 return r;
4205 break;
4206
4207 case ARG_FACTORY_RESET:
4208 r = parse_boolean_argument("--factory-reset=", optarg, NULL);
4209 if (r < 0)
4210 return r;
4211 arg_factory_reset = r;
4212 break;
4213
4214 case ARG_CAN_FACTORY_RESET:
4215 arg_can_factory_reset = true;
4216 break;
4217
4218 case ARG_ROOT:
4219 r = parse_path_argument(optarg, /* suppress_root= */ false, &arg_root);
4220 if (r < 0)
4221 return r;
4222 break;
4223
4224 case ARG_IMAGE:
4225 r = parse_path_argument(optarg, /* suppress_root= */ false, &arg_image);
4226 if (r < 0)
4227 return r;
4228 break;
4229
4230 case ARG_SEED:
4231 if (isempty(optarg)) {
4232 arg_seed = SD_ID128_NULL;
4233 arg_randomize = false;
4234 } else if (streq(optarg, "random"))
4235 arg_randomize = true;
4236 else {
4237 r = sd_id128_from_string(optarg, &arg_seed);
4238 if (r < 0)
4239 return log_error_errno(r, "Failed to parse seed: %s", optarg);
4240
4241 arg_randomize = false;
4242 }
4243
4244 break;
4245
4246 case ARG_PRETTY:
4247 r = parse_boolean_argument("--pretty=", optarg, NULL);
4248 if (r < 0)
4249 return r;
4250 arg_pretty = r;
4251 break;
4252
4253 case ARG_DEFINITIONS:
4254 r = parse_path_argument(optarg, false, &arg_definitions);
4255 if (r < 0)
4256 return r;
4257 break;
4258
4259 case ARG_SIZE: {
4260 uint64_t parsed, rounded;
4261
4262 if (streq(optarg, "auto")) {
4263 arg_size = UINT64_MAX;
4264 arg_size_auto = true;
4265 break;
4266 }
4267
4268 r = parse_size(optarg, 1024, &parsed);
4269 if (r < 0)
4270 return log_error_errno(r, "Failed to parse --size= parameter: %s", optarg);
4271
4272 rounded = round_up_size(parsed, 4096);
4273 if (rounded == 0)
4274 return log_error_errno(SYNTHETIC_ERRNO(ERANGE), "Specified image size too small, refusing.");
4275 if (rounded == UINT64_MAX)
4276 return log_error_errno(SYNTHETIC_ERRNO(ERANGE), "Specified image size too large, refusing.");
4277
4278 if (rounded != parsed)
4279 log_warning("Specified size is not a multiple of 4096, rounding up automatically. (%" PRIu64 " → %" PRIu64 ")",
4280 parsed, rounded);
4281
4282 arg_size = rounded;
4283 arg_size_auto = false;
4284 break;
4285 }
4286
4287 case ARG_JSON:
4288 r = parse_json_argument(optarg, &arg_json_format_flags);
4289 if (r <= 0)
4290 return r;
4291
4292 break;
4293
4294 case ARG_KEY_FILE: {
4295 _cleanup_(erase_and_freep) char *k = NULL;
4296 size_t n = 0;
4297
4298 r = read_full_file_full(
4299 AT_FDCWD, optarg, UINT64_MAX, SIZE_MAX,
4300 READ_FULL_FILE_SECURE|READ_FULL_FILE_WARN_WORLD_READABLE|READ_FULL_FILE_CONNECT_SOCKET,
4301 NULL,
4302 &k, &n);
4303 if (r < 0)
4304 return log_error_errno(r, "Failed to read key file '%s': %m", optarg);
4305
4306 erase_and_free(arg_key);
4307 arg_key = TAKE_PTR(k);
4308 arg_key_size = n;
4309 break;
4310 }
4311
4312 case ARG_TPM2_DEVICE: {
4313 _cleanup_free_ char *device = NULL;
4314
4315 if (streq(optarg, "list"))
4316 return tpm2_list_devices();
4317
4318 if (!streq(optarg, "auto")) {
4319 device = strdup(optarg);
4320 if (!device)
4321 return log_oom();
4322 }
4323
4324 free(arg_tpm2_device);
4325 arg_tpm2_device = TAKE_PTR(device);
4326 break;
4327 }
4328
4329 case ARG_TPM2_PCRS: {
4330 uint32_t mask;
4331
4332 if (isempty(optarg)) {
4333 arg_tpm2_pcr_mask = 0;
4334 break;
4335 }
4336
4337 r = tpm2_parse_pcrs(optarg, &mask);
4338 if (r < 0)
4339 return r;
4340
4341 if (arg_tpm2_pcr_mask == UINT32_MAX)
4342 arg_tpm2_pcr_mask = mask;
4343 else
4344 arg_tpm2_pcr_mask |= mask;
4345
4346 break;
4347 }
4348
4349 case '?':
4350 return -EINVAL;
4351
4352 default:
4353 assert_not_reached();
4354 }
4355
4356 if (argc - optind > 1)
4357 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4358 "Expected at most one argument, the path to the block device.");
4359
4360 if (arg_factory_reset > 0 && IN_SET(arg_empty, EMPTY_FORCE, EMPTY_REQUIRE, EMPTY_CREATE))
4361 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4362 "Combination of --factory-reset=yes and --empty=force/--empty=require/--empty=create is invalid.");
4363
4364 if (arg_can_factory_reset)
4365 arg_dry_run = true; /* When --can-factory-reset is specified we don't make changes, hence
4366 * non-dry-run mode makes no sense. Thus, imply dry run mode so that we
4367 * open things strictly read-only. */
4368 else if (dry_run >= 0)
4369 arg_dry_run = dry_run;
4370
4371 if (arg_empty == EMPTY_CREATE && (arg_size == UINT64_MAX && !arg_size_auto))
4372 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4373 "If --empty=create is specified, --size= must be specified, too.");
4374
4375 if (arg_image && arg_root)
4376 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Please specify either --root= or --image=, the combination of both is not supported.");
4377 else if (!arg_image && !arg_root && in_initrd()) {
4378
4379 /* By default operate on /sysusr/ or /sysroot/ when invoked in the initrd. We prefer the
4380 * former, if it is mounted, so that we have deterministic behaviour on systems where /usr/
4381 * is vendor-supplied but the root fs formatted on first boot. */
4382 r = path_is_mount_point("/sysusr/usr", NULL, 0);
4383 if (r <= 0) {
4384 if (r < 0 && r != -ENOENT)
4385 log_debug_errno(r, "Unable to determine whether /sysusr/usr is a mount point, assuming it is not: %m");
4386
4387 arg_root = strdup("/sysroot");
4388 } else
4389 arg_root = strdup("/sysusr");
4390 if (!arg_root)
4391 return log_oom();
4392 }
4393
4394 arg_node = argc > optind ? argv[optind] : NULL;
4395
4396 if (IN_SET(arg_empty, EMPTY_FORCE, EMPTY_REQUIRE, EMPTY_CREATE) && !arg_node && !arg_image)
4397 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4398 "A path to a device node or loopback file must be specified when --empty=force, --empty=require or --empty=create are used.");
4399
4400 if (arg_tpm2_pcr_mask == UINT32_MAX)
4401 arg_tpm2_pcr_mask = TPM2_PCR_MASK_DEFAULT;
4402
4403 return 1;
4404 }
4405
4406 static int parse_proc_cmdline_factory_reset(void) {
4407 bool b;
4408 int r;
4409
4410 if (arg_factory_reset >= 0) /* Never override what is specified on the process command line */
4411 return 0;
4412
4413 if (!in_initrd()) /* Never honour kernel command line factory reset request outside of the initrd */
4414 return 0;
4415
4416 r = proc_cmdline_get_bool("systemd.factory_reset", &b);
4417 if (r < 0)
4418 return log_error_errno(r, "Failed to parse systemd.factory_reset kernel command line argument: %m");
4419 if (r > 0) {
4420 arg_factory_reset = b;
4421
4422 if (b)
4423 log_notice("Honouring factory reset requested via kernel command line.");
4424 }
4425
4426 return 0;
4427 }
4428
4429 static int parse_efi_variable_factory_reset(void) {
4430 _cleanup_free_ char *value = NULL;
4431 int r;
4432
4433 if (arg_factory_reset >= 0) /* Never override what is specified on the process command line */
4434 return 0;
4435
4436 if (!in_initrd()) /* Never honour EFI variable factory reset request outside of the initrd */
4437 return 0;
4438
4439 r = efi_get_variable_string(EFI_SYSTEMD_VARIABLE(FactoryReset), &value);
4440 if (r == -ENOENT || ERRNO_IS_NOT_SUPPORTED(r))
4441 return 0;
4442 if (r < 0)
4443 return log_error_errno(r, "Failed to read EFI variable FactoryReset: %m");
4444
4445 r = parse_boolean(value);
4446 if (r < 0)
4447 return log_error_errno(r, "Failed to parse EFI variable FactoryReset: %m");
4448
4449 arg_factory_reset = r;
4450 if (r)
4451 log_notice("Factory reset requested via EFI variable FactoryReset.");
4452
4453 return 0;
4454 }
4455
4456 static int remove_efi_variable_factory_reset(void) {
4457 int r;
4458
4459 r = efi_set_variable(EFI_SYSTEMD_VARIABLE(FactoryReset), NULL, 0);
4460 if (r == -ENOENT || ERRNO_IS_NOT_SUPPORTED(r))
4461 return 0;
4462 if (r < 0)
4463 return log_error_errno(r, "Failed to remove EFI variable FactoryReset: %m");
4464
4465 log_info("Successfully unset EFI variable FactoryReset.");
4466 return 0;
4467 }
4468
4469 static int acquire_root_devno(
4470 const char *p,
4471 const char *root,
4472 int mode,
4473 char **ret,
4474 int *ret_fd) {
4475
4476 _cleanup_free_ char *found_path = NULL;
4477 dev_t devno, fd_devno = MODE_INVALID;
4478 _cleanup_close_ int fd = -1;
4479 struct stat st;
4480 int r;
4481
4482 assert(p);
4483 assert(ret);
4484 assert(ret_fd);
4485
4486 fd = chase_symlinks_and_open(p, root, CHASE_PREFIX_ROOT, mode, &found_path);
4487 if (fd < 0)
4488 return fd;
4489
4490 if (fstat(fd, &st) < 0)
4491 return -errno;
4492
4493 if (S_ISREG(st.st_mode)) {
4494 *ret = TAKE_PTR(found_path);
4495 *ret_fd = TAKE_FD(fd);
4496 return 0;
4497 }
4498
4499 if (S_ISBLK(st.st_mode)) {
4500 /* Refuse referencing explicit block devices if a root dir is specified, after all we should
4501 * not be able to leave the image the root path constrains us to. */
4502 if (root)
4503 return -EPERM;
4504
4505 fd_devno = devno = st.st_rdev;
4506 } else if (S_ISDIR(st.st_mode)) {
4507
4508 devno = st.st_dev;
4509 if (major(devno) == 0) {
4510 r = btrfs_get_block_device_fd(fd, &devno);
4511 if (r == -ENOTTY) /* not btrfs */
4512 return -ENODEV;
4513 if (r < 0)
4514 return r;
4515 }
4516 } else
4517 return -ENOTBLK;
4518
4519 /* From dm-crypt to backing partition */
4520 r = block_get_originating(devno, &devno);
4521 if (r == -ENOENT)
4522 log_debug_errno(r, "Device '%s' has no dm-crypt/dm-verity device, no need to look for underlying block device.", p);
4523 else if (r < 0)
4524 log_debug_errno(r, "Failed to find underlying block device for '%s', ignoring: %m", p);
4525
4526 /* From partition to whole disk containing it */
4527 r = block_get_whole_disk(devno, &devno);
4528 if (r < 0)
4529 log_debug_errno(r, "Failed to find whole disk block device for '%s', ignoring: %m", p);
4530
4531 r = device_path_make_canonical(S_IFBLK, devno, ret);
4532 if (r < 0)
4533 return log_debug_errno(r, "Failed to determine canonical path for '%s': %m", p);
4534
4535 /* Only if we still look at the same block device we can reuse the fd. Otherwise return an
4536 * invalidated fd. */
4537 *ret_fd = fd_devno != MODE_INVALID && fd_devno == devno ? TAKE_FD(fd) : -1;
4538 return 0;
4539 }
4540
4541 static int find_root(char **ret, int *ret_fd) {
4542 _cleanup_free_ char *device = NULL;
4543 int r;
4544
4545 assert(ret);
4546 assert(ret_fd);
4547
4548 if (arg_node) {
4549 if (arg_empty == EMPTY_CREATE) {
4550 _cleanup_close_ int fd = -1;
4551 _cleanup_free_ char *s = NULL;
4552
4553 s = strdup(arg_node);
4554 if (!s)
4555 return log_oom();
4556
4557 fd = open(arg_node, O_RDONLY|O_CREAT|O_EXCL|O_CLOEXEC|O_NOFOLLOW, 0666);
4558 if (fd < 0)
4559 return log_error_errno(errno, "Failed to create '%s': %m", arg_node);
4560
4561 *ret = TAKE_PTR(s);
4562 *ret_fd = TAKE_FD(fd);
4563 return 0;
4564 }
4565
4566 /* Note that we don't specify a root argument here: if the user explicitly configured a node
4567 * we'll take it relative to the host, not the image */
4568 r = acquire_root_devno(arg_node, NULL, O_RDONLY|O_CLOEXEC, ret, ret_fd);
4569 if (r == -EUCLEAN)
4570 return btrfs_log_dev_root(LOG_ERR, r, arg_node);
4571 if (r < 0)
4572 return log_error_errno(r, "Failed to open file or determine backing device of %s: %m", arg_node);
4573
4574 return 0;
4575 }
4576
4577 assert(IN_SET(arg_empty, EMPTY_REFUSE, EMPTY_ALLOW));
4578
4579 /* If the root mount has been replaced by some form of volatile file system (overlayfs), the
4580 * original root block device node is symlinked in /run/systemd/volatile-root. Let's read that
4581 * here. */
4582 r = readlink_malloc("/run/systemd/volatile-root", &device);
4583 if (r == -ENOENT) { /* volatile-root not found */
4584 /* Let's search for the root device. We look for two cases here: first in /, and then in /usr. The
4585 * latter we check for cases where / is a tmpfs and only /usr is an actual persistent block device
4586 * (think: volatile setups) */
4587
4588 FOREACH_STRING(p, "/", "/usr") {
4589
4590 r = acquire_root_devno(p, arg_root, O_RDONLY|O_DIRECTORY|O_CLOEXEC, ret, ret_fd);
4591 if (r < 0) {
4592 if (r == -EUCLEAN)
4593 return btrfs_log_dev_root(LOG_ERR, r, p);
4594 if (r != -ENODEV)
4595 return log_error_errno(r, "Failed to determine backing device of %s: %m", p);
4596 } else
4597 return 0;
4598 }
4599 } else if (r < 0)
4600 return log_error_errno(r, "Failed to read symlink /run/systemd/volatile-root: %m");
4601 else {
4602 r = acquire_root_devno(device, NULL, O_RDONLY|O_CLOEXEC, ret, ret_fd);
4603 if (r == -EUCLEAN)
4604 return btrfs_log_dev_root(LOG_ERR, r, device);
4605 if (r < 0)
4606 return log_error_errno(r, "Failed to open file or determine backing device of %s: %m", device);
4607
4608 return 0;
4609 }
4610
4611 return log_error_errno(SYNTHETIC_ERRNO(ENODEV), "Failed to discover root block device.");
4612 }
4613
4614 static int resize_pt(int fd) {
4615 _cleanup_(fdisk_unref_contextp) struct fdisk_context *c = NULL;
4616 int r;
4617
4618 /* After resizing the backing file we need to resize the partition table itself too, so that it takes
4619 * possession of the enlarged backing file. For this it suffices to open the device with libfdisk and
4620 * immediately write it again, with no changes. */
4621
4622 c = fdisk_new_context();
4623 if (!c)
4624 return log_oom();
4625
4626 r = fdisk_assign_device(c, FORMAT_PROC_FD_PATH(fd), 0);
4627 if (r < 0)
4628 return log_error_errno(r, "Failed to open device '%s': %m", FORMAT_PROC_FD_PATH(fd));
4629
4630 r = fdisk_has_label(c);
4631 if (r < 0)
4632 return log_error_errno(r, "Failed to determine whether disk '%s' has a disk label: %m", FORMAT_PROC_FD_PATH(fd));
4633 if (r == 0) {
4634 log_debug("Not resizing partition table, as there currently is none.");
4635 return 0;
4636 }
4637
4638 r = fdisk_write_disklabel(c);
4639 if (r < 0)
4640 return log_error_errno(r, "Failed to write resized partition table: %m");
4641
4642 log_info("Resized partition table.");
4643 return 1;
4644 }
4645
4646 static int resize_backing_fd(
4647 const char *node, /* The primary way we access the disk image to operate on */
4648 int *fd, /* An O_RDONLY fd referring to that inode */
4649 const char *backing_file, /* If the above refers to a loopback device, the backing regular file for that, which we can grow */
4650 LoopDevice *loop_device) {
4651
4652 _cleanup_close_ int writable_fd = -1;
4653 uint64_t current_size;
4654 struct stat st;
4655 int r;
4656
4657 assert(node);
4658 assert(fd);
4659
4660 if (arg_size == UINT64_MAX) /* Nothing to do */
4661 return 0;
4662
4663 if (*fd < 0) {
4664 /* Open the file if we haven't opened it yet. Note that we open it read-only here, just to
4665 * keep a reference to the file we can pass around. */
4666 *fd = open(node, O_RDONLY|O_CLOEXEC);
4667 if (*fd < 0)
4668 return log_error_errno(errno, "Failed to open '%s' in order to adjust size: %m", node);
4669 }
4670
4671 if (fstat(*fd, &st) < 0)
4672 return log_error_errno(errno, "Failed to stat '%s': %m", node);
4673
4674 if (S_ISBLK(st.st_mode)) {
4675 if (!backing_file)
4676 return log_error_errno(SYNTHETIC_ERRNO(EBADF), "Cannot resize block device '%s'.", node);
4677
4678 assert(loop_device);
4679
4680 if (ioctl(*fd, BLKGETSIZE64, &current_size) < 0)
4681 return log_error_errno(errno, "Failed to determine size of block device %s: %m", node);
4682 } else {
4683 r = stat_verify_regular(&st);
4684 if (r < 0)
4685 return log_error_errno(r, "Specified path '%s' is not a regular file or loopback block device, cannot resize: %m", node);
4686
4687 assert(!backing_file);
4688 assert(!loop_device);
4689 current_size = st.st_size;
4690 }
4691
4692 if (current_size >= arg_size) {
4693 log_info("File '%s' already is of requested size or larger, not growing. (%s >= %s)",
4694 node, FORMAT_BYTES(current_size), FORMAT_BYTES(arg_size));
4695 return 0;
4696 }
4697
4698 if (S_ISBLK(st.st_mode)) {
4699 assert(backing_file);
4700
4701 /* This is a loopback device. We can't really grow those directly, but we can grow the
4702 * backing file, hence let's do that. */
4703
4704 writable_fd = open(backing_file, O_WRONLY|O_CLOEXEC|O_NONBLOCK);
4705 if (writable_fd < 0)
4706 return log_error_errno(errno, "Failed to open backing file '%s': %m", backing_file);
4707
4708 if (fstat(writable_fd, &st) < 0)
4709 return log_error_errno(errno, "Failed to stat() backing file '%s': %m", backing_file);
4710
4711 r = stat_verify_regular(&st);
4712 if (r < 0)
4713 return log_error_errno(r, "Backing file '%s' of block device is not a regular file: %m", backing_file);
4714
4715 if ((uint64_t) st.st_size != current_size)
4716 return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
4717 "Size of backing file '%s' of loopback block device '%s' don't match, refusing.",
4718 node, backing_file);
4719 } else {
4720 assert(S_ISREG(st.st_mode));
4721 assert(!backing_file);
4722
4723 /* The file descriptor is read-only. In order to grow the file we need to have a writable fd. We
4724 * reopen the file for that temporarily. We keep the writable fd only open for this operation though,
4725 * as fdisk can't accept it anyway. */
4726
4727 writable_fd = fd_reopen(*fd, O_WRONLY|O_CLOEXEC);
4728 if (writable_fd < 0)
4729 return log_error_errno(writable_fd, "Failed to reopen backing file '%s' writable: %m", node);
4730 }
4731
4732 if (!arg_discard) {
4733 if (fallocate(writable_fd, 0, 0, arg_size) < 0) {
4734 if (!ERRNO_IS_NOT_SUPPORTED(errno))
4735 return log_error_errno(errno, "Failed to grow '%s' from %s to %s by allocation: %m",
4736 node, FORMAT_BYTES(current_size), FORMAT_BYTES(arg_size));
4737
4738 /* Fallback to truncation, if fallocate() is not supported. */
4739 log_debug("Backing file system does not support fallocate(), falling back to ftruncate().");
4740 } else {
4741 if (current_size == 0) /* Likely regular file just created by us */
4742 log_info("Allocated %s for '%s'.", FORMAT_BYTES(arg_size), node);
4743 else
4744 log_info("File '%s' grown from %s to %s by allocation.",
4745 node, FORMAT_BYTES(current_size), FORMAT_BYTES(arg_size));
4746
4747 goto done;
4748 }
4749 }
4750
4751 if (ftruncate(writable_fd, arg_size) < 0)
4752 return log_error_errno(errno, "Failed to grow '%s' from %s to %s by truncation: %m",
4753 node, FORMAT_BYTES(current_size), FORMAT_BYTES(arg_size));
4754
4755 if (current_size == 0) /* Likely regular file just created by us */
4756 log_info("Sized '%s' to %s.", node, FORMAT_BYTES(arg_size));
4757 else
4758 log_info("File '%s' grown from %s to %s by truncation.",
4759 node, FORMAT_BYTES(current_size), FORMAT_BYTES(arg_size));
4760
4761 done:
4762 r = resize_pt(writable_fd);
4763 if (r < 0)
4764 return r;
4765
4766 if (loop_device) {
4767 r = loop_device_refresh_size(loop_device, UINT64_MAX, arg_size);
4768 if (r < 0)
4769 return log_error_errno(r, "Failed to update loop device size: %m");
4770 }
4771
4772 return 1;
4773 }
4774
4775 static int determine_auto_size(Context *c) {
4776 uint64_t sum;
4777
4778 assert(c);
4779
4780 sum = round_up_size(GPT_METADATA_SIZE, 4096);
4781
4782 LIST_FOREACH(partitions, p, c->partitions) {
4783 uint64_t m;
4784
4785 if (p->dropped)
4786 continue;
4787
4788 m = partition_min_size_with_padding(c, p);
4789 if (m > UINT64_MAX - sum)
4790 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Image would grow too large, refusing.");
4791
4792 sum += m;
4793 }
4794
4795 if (c->total != UINT64_MAX)
4796 /* Image already allocated? Then show its size. */
4797 log_info("Automatically determined minimal disk image size as %s, current image size is %s.",
4798 FORMAT_BYTES(sum), FORMAT_BYTES(c->total));
4799 else
4800 /* If the image is being created right now, then it has no previous size, suppress any comment about it hence. */
4801 log_info("Automatically determined minimal disk image size as %s.",
4802 FORMAT_BYTES(sum));
4803
4804 arg_size = sum;
4805 return 0;
4806 }
4807
4808 static int run(int argc, char *argv[]) {
4809 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
4810 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
4811 _cleanup_(umount_and_rmdir_and_freep) char *mounted_dir = NULL;
4812 _cleanup_(context_freep) Context* context = NULL;
4813 _cleanup_free_ char *node = NULL;
4814 _cleanup_close_ int backing_fd = -1;
4815 bool from_scratch, node_is_our_loop = false;
4816 int r;
4817
4818 log_show_color(true);
4819 log_parse_environment();
4820 log_open();
4821
4822 r = parse_argv(argc, argv);
4823 if (r <= 0)
4824 return r;
4825
4826 r = parse_proc_cmdline_factory_reset();
4827 if (r < 0)
4828 return r;
4829
4830 r = parse_efi_variable_factory_reset();
4831 if (r < 0)
4832 return r;
4833
4834 #if HAVE_LIBCRYPTSETUP
4835 cryptsetup_enable_logging(NULL);
4836 #endif
4837
4838 if (arg_image) {
4839 assert(!arg_root);
4840
4841 /* Mount this strictly read-only: we shall modify the partition table, not the file
4842 * systems */
4843 r = mount_image_privately_interactively(
4844 arg_image,
4845 DISSECT_IMAGE_MOUNT_READ_ONLY |
4846 (arg_node ? DISSECT_IMAGE_DEVICE_READ_ONLY : 0) | /* If a different node to make changes to is specified let's open the device in read-only mode) */
4847 DISSECT_IMAGE_GPT_ONLY |
4848 DISSECT_IMAGE_RELAX_VAR_CHECK |
4849 DISSECT_IMAGE_USR_NO_ROOT |
4850 DISSECT_IMAGE_REQUIRE_ROOT,
4851 &mounted_dir,
4852 &loop_device,
4853 &decrypted_image);
4854 if (r < 0)
4855 return r;
4856
4857 arg_root = strdup(mounted_dir);
4858 if (!arg_root)
4859 return log_oom();
4860
4861 if (!arg_node) {
4862 arg_node = strdup(loop_device->node);
4863 if (!arg_node)
4864 return log_oom();
4865
4866 /* Remember that the device we are about to manipulate is actually the one we
4867 * allocated here, and thus to increase its backing file we know what to do */
4868 node_is_our_loop = true;
4869 }
4870 }
4871
4872 context = context_new(arg_seed);
4873 if (!context)
4874 return log_oom();
4875
4876 r = context_read_definitions(context, arg_definitions, arg_root);
4877 if (r < 0)
4878 return r;
4879
4880 if (context->n_partitions <= 0 && arg_empty == EMPTY_REFUSE) {
4881 log_info("Didn't find any partition definition files, nothing to do.");
4882 return 0;
4883 }
4884
4885 r = find_root(&node, &backing_fd);
4886 if (r < 0)
4887 return r;
4888
4889 if (arg_size != UINT64_MAX) {
4890 r = resize_backing_fd(
4891 node,
4892 &backing_fd,
4893 node_is_our_loop ? arg_image : NULL,
4894 node_is_our_loop ? loop_device : NULL);
4895 if (r < 0)
4896 return r;
4897 }
4898
4899 r = context_load_partition_table(context, node, &backing_fd);
4900 if (r == -EHWPOISON)
4901 return 77; /* Special return value which means "Not GPT, so not doing anything". This isn't
4902 * really an error when called at boot. */
4903 if (r < 0)
4904 return r;
4905 from_scratch = r > 0; /* Starting from scratch */
4906
4907 if (arg_can_factory_reset) {
4908 r = context_can_factory_reset(context);
4909 if (r < 0)
4910 return r;
4911 if (r == 0)
4912 return EXIT_FAILURE;
4913
4914 return 0;
4915 }
4916
4917 r = context_factory_reset(context, from_scratch);
4918 if (r < 0)
4919 return r;
4920 if (r > 0) {
4921 /* We actually did a factory reset! */
4922 r = remove_efi_variable_factory_reset();
4923 if (r < 0)
4924 return r;
4925
4926 /* Reload the reduced partition table */
4927 context_unload_partition_table(context);
4928 r = context_load_partition_table(context, node, &backing_fd);
4929 if (r < 0)
4930 return r;
4931 }
4932
4933 #if 0
4934 (void) context_dump_partitions(context, node);
4935 putchar('\n');
4936 #endif
4937
4938 r = context_read_seed(context, arg_root);
4939 if (r < 0)
4940 return r;
4941
4942 /* Open all files to copy blocks from now, since we want to take their size into consideration */
4943 r = context_open_copy_block_paths(
4944 context,
4945 arg_root,
4946 loop_device ? loop_device->devno : /* if --image= is specified, only allow partitions on the loopback device */
4947 arg_root && !arg_image ? 0 : /* if --root= is specified, don't accept any block device */
4948 (dev_t) -1); /* if neither is specified, make no restrictions */
4949 if (r < 0)
4950 return r;
4951
4952 if (arg_size_auto) {
4953 r = determine_auto_size(context);
4954 if (r < 0)
4955 return r;
4956
4957 /* Flush out everything again, and let's grow the file first, then start fresh */
4958 context_unload_partition_table(context);
4959
4960 assert(arg_size != UINT64_MAX);
4961 r = resize_backing_fd(
4962 node,
4963 &backing_fd,
4964 node_is_our_loop ? arg_image : NULL,
4965 node_is_our_loop ? loop_device : NULL);
4966 if (r < 0)
4967 return r;
4968
4969 r = context_load_partition_table(context, node, &backing_fd);
4970 if (r < 0)
4971 return r;
4972 }
4973
4974 /* First try to fit new partitions in, dropping by priority until it fits */
4975 for (;;) {
4976 uint64_t largest_free_area;
4977
4978 if (context_allocate_partitions(context, &largest_free_area))
4979 break; /* Success! */
4980
4981 if (!context_drop_one_priority(context)) {
4982 r = log_error_errno(SYNTHETIC_ERRNO(ENOSPC),
4983 "Can't fit requested partitions into available free space (%s), refusing.",
4984 FORMAT_BYTES(largest_free_area));
4985 determine_auto_size(context);
4986 return r;
4987 }
4988 }
4989
4990 /* Now assign free space according to the weight logic */
4991 r = context_grow_partitions(context);
4992 if (r < 0)
4993 return r;
4994
4995 /* Now calculate where each new partition gets placed */
4996 context_place_partitions(context);
4997
4998 /* Make sure each partition has a unique UUID and unique label */
4999 r = context_acquire_partition_uuids_and_labels(context);
5000 if (r < 0)
5001 return r;
5002
5003 r = context_write_partition_table(context, node, from_scratch);
5004 if (r < 0)
5005 return r;
5006
5007 return 0;
5008 }
5009
5010 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
Unnamed repository; edit this file 'description' to name the repository.