Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * kaslr.c
4 *
5 * This contains the routines needed to generate a reasonable level of
6 * entropy to choose a randomized kernel base address offset in support
7 * of Kernel Address Space Layout Randomization (KASLR). Additionally
8 * handles walking the physical memory maps (and tracking memory regions
9 * to avoid) in order to select a physical memory location that can
10 * contain the entire properly aligned running kernel image.
11 *
12 */
13
14/*
15 * isspace() in linux/ctype.h is expected by next_args() to filter
16 * out "space/lf/tab". While boot/ctype.h conflicts with linux/ctype.h,
17 * since isdigit() is implemented in both of them. Hence disable it
18 * here.
19 */
20#define BOOT_CTYPE_H
21
22#include "misc.h"
23#include "error.h"
24#include "../string.h"
25#include "efi.h"
26
27#include <generated/compile.h>
28#include <linux/module.h>
29#include <linux/uts.h>
30#include <linux/utsname.h>
31#include <linux/ctype.h>
32#include <generated/utsversion.h>
33#include <generated/utsrelease.h>
34
35#define _SETUP
36#include <asm/setup.h> /* For COMMAND_LINE_SIZE */
37#undef _SETUP
38
39extern unsigned long get_cmd_line_ptr(void);
40
41/* Simplified build-specific string for starting entropy. */
42static const char build_str[] = UTS_RELEASE " (" LINUX_COMPILE_BY "@"
43 LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION;
44
45static unsigned long rotate_xor(unsigned long hash, const void *area,
46 size_t size)
47{
48 size_t i;
49 unsigned long *ptr = (unsigned long *)area;
50
51 for (i = 0; i < size / sizeof(hash); i++) {
52 /* Rotate by odd number of bits and XOR. */
53 hash = (hash << ((sizeof(hash) * 8) - 7)) | (hash >> 7);
54 hash ^= ptr[i];
55 }
56
57 return hash;
58}
59
60/* Attempt to create a simple but unpredictable starting entropy. */
61static unsigned long get_boot_seed(void)
62{
63 unsigned long hash = 0;
64
65 hash = rotate_xor(hash, build_str, sizeof(build_str));
66 hash = rotate_xor(hash, boot_params_ptr, sizeof(*boot_params_ptr));
67
68 return hash;
69}
70
71#define KASLR_COMPRESSED_BOOT
72#include "../../lib/kaslr.c"
73
74
75/* Only supporting at most 4 unusable memmap regions with kaslr */
76#define MAX_MEMMAP_REGIONS 4
77
78static bool memmap_too_large;
79
80
81/*
82 * Store memory limit: MAXMEM on 64-bit and KERNEL_IMAGE_SIZE on 32-bit.
83 * It may be reduced by "mem=nn[KMG]" or "memmap=nn[KMG]" command line options.
84 */
85static u64 mem_limit;
86
87/* Number of immovable memory regions */
88static int num_immovable_mem;
89
90enum mem_avoid_index {
91 MEM_AVOID_ZO_RANGE = 0,
92 MEM_AVOID_INITRD,
93 MEM_AVOID_CMDLINE,
94 MEM_AVOID_BOOTPARAMS,
95 MEM_AVOID_MEMMAP_BEGIN,
96 MEM_AVOID_MEMMAP_END = MEM_AVOID_MEMMAP_BEGIN + MAX_MEMMAP_REGIONS - 1,
97 MEM_AVOID_MAX,
98};
99
100static struct mem_vector mem_avoid[MEM_AVOID_MAX];
101
102static bool mem_overlaps(struct mem_vector *one, struct mem_vector *two)
103{
104 /* Item one is entirely before item two. */
105 if (one->start + one->size <= two->start)
106 return false;
107 /* Item one is entirely after item two. */
108 if (one->start >= two->start + two->size)
109 return false;
110 return true;
111}
112
113char *skip_spaces(const char *str)
114{
115 while (isspace(*str))
116 ++str;
117 return (char *)str;
118}
119#include "../../../../lib/ctype.c"
120#include "../../../../lib/cmdline.c"
121
122static int
123parse_memmap(char *p, u64 *start, u64 *size)
124{
125 char *oldp;
126
127 if (!p)
128 return -EINVAL;
129
130 /* We don't care about this option here */
131 if (!strncmp(p, "exactmap", 8))
132 return -EINVAL;
133
134 oldp = p;
135 *size = memparse(p, &p);
136 if (p == oldp)
137 return -EINVAL;
138
139 switch (*p) {
140 case '#':
141 case '$':
142 case '!':
143 *start = memparse(p + 1, &p);
144 return 0;
145 case '@':
146 /*
147 * memmap=nn@ss specifies usable region, should
148 * be skipped
149 */
150 *size = 0;
151 fallthrough;
152 default:
153 /*
154 * If w/o offset, only size specified, memmap=nn[KMG] has the
155 * same behaviour as mem=nn[KMG]. It limits the max address
156 * system can use. Region above the limit should be avoided.
157 */
158 *start = 0;
159 return 0;
160 }
161
162 return -EINVAL;
163}
164
165static void mem_avoid_memmap(char *str)
166{
167 static int i;
168
169 if (i >= MAX_MEMMAP_REGIONS)
170 return;
171
172 while (str && (i < MAX_MEMMAP_REGIONS)) {
173 int rc;
174 u64 start, size;
175 char *k = strchr(str, ',');
176
177 if (k)
178 *k++ = 0;
179
180 rc = parse_memmap(str, &start, &size);
181 if (rc < 0)
182 break;
183 str = k;
184
185 if (start == 0) {
186 /* Store the specified memory limit if size > 0 */
187 if (size > 0 && size < mem_limit)
188 mem_limit = size;
189
190 continue;
191 }
192
193 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
194 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
195 i++;
196 }
197
198 /* More than 4 memmaps, fail kaslr */
199 if ((i >= MAX_MEMMAP_REGIONS) && str)
200 memmap_too_large = true;
201}
202
203/* Store the number of 1GB huge pages which users specified: */
204static unsigned long max_gb_huge_pages;
205
206static void parse_gb_huge_pages(char *param, char *val)
207{
208 static bool gbpage_sz;
209 char *p;
210
211 if (!strcmp(param, "hugepagesz")) {
212 p = val;
213 if (memparse(p, &p) != PUD_SIZE) {
214 gbpage_sz = false;
215 return;
216 }
217
218 if (gbpage_sz)
219 warn("Repeatedly set hugeTLB page size of 1G!\n");
220 gbpage_sz = true;
221 return;
222 }
223
224 if (!strcmp(param, "hugepages") && gbpage_sz) {
225 p = val;
226 max_gb_huge_pages = simple_strtoull(p, &p, 0);
227 return;
228 }
229}
230
231static void handle_mem_options(void)
232{
233 char *args = (char *)get_cmd_line_ptr();
234 size_t len;
235 char *tmp_cmdline;
236 char *param, *val;
237 u64 mem_size;
238
239 if (!args)
240 return;
241
242 len = strnlen(args, COMMAND_LINE_SIZE-1);
243 tmp_cmdline = malloc(len + 1);
244 if (!tmp_cmdline)
245 error("Failed to allocate space for tmp_cmdline");
246
247 memcpy(tmp_cmdline, args, len);
248 tmp_cmdline[len] = 0;
249 args = tmp_cmdline;
250
251 /* Chew leading spaces */
252 args = skip_spaces(args);
253
254 while (*args) {
255 args = next_arg(args, ¶m, &val);
256 /* Stop at -- */
257 if (!val && strcmp(param, "--") == 0)
258 break;
259
260 if (!strcmp(param, "memmap")) {
261 mem_avoid_memmap(val);
262 } else if (IS_ENABLED(CONFIG_X86_64) && strstr(param, "hugepages")) {
263 parse_gb_huge_pages(param, val);
264 } else if (!strcmp(param, "mem")) {
265 char *p = val;
266
267 if (!strcmp(p, "nopentium"))
268 continue;
269 mem_size = memparse(p, &p);
270 if (mem_size == 0)
271 break;
272
273 if (mem_size < mem_limit)
274 mem_limit = mem_size;
275 }
276 }
277
278 free(tmp_cmdline);
279 return;
280}
281
282/*
283 * In theory, KASLR can put the kernel anywhere in the range of [16M, MAXMEM)
284 * on 64-bit, and [16M, KERNEL_IMAGE_SIZE) on 32-bit.
285 *
286 * The mem_avoid array is used to store the ranges that need to be avoided
287 * when KASLR searches for an appropriate random address. We must avoid any
288 * regions that are unsafe to overlap with during decompression, and other
289 * things like the initrd, cmdline and boot_params. This comment seeks to
290 * explain mem_avoid as clearly as possible since incorrect mem_avoid
291 * memory ranges lead to really hard to debug boot failures.
292 *
293 * The initrd, cmdline, and boot_params are trivial to identify for
294 * avoiding. They are MEM_AVOID_INITRD, MEM_AVOID_CMDLINE, and
295 * MEM_AVOID_BOOTPARAMS respectively below.
296 *
297 * What is not obvious how to avoid is the range of memory that is used
298 * during decompression (MEM_AVOID_ZO_RANGE below). This range must cover
299 * the compressed kernel (ZO) and its run space, which is used to extract
300 * the uncompressed kernel (VO) and relocs.
301 *
302 * ZO's full run size sits against the end of the decompression buffer, so
303 * we can calculate where text, data, bss, etc of ZO are positioned more
304 * easily.
305 *
306 * For additional background, the decompression calculations can be found
307 * in header.S, and the memory diagram is based on the one found in misc.c.
308 *
309 * The following conditions are already enforced by the image layouts and
310 * associated code:
311 * - input + input_size >= output + output_size
312 * - kernel_total_size <= init_size
313 * - kernel_total_size <= output_size (see Note below)
314 * - output + init_size >= output + output_size
315 *
316 * (Note that kernel_total_size and output_size have no fundamental
317 * relationship, but output_size is passed to choose_random_location
318 * as a maximum of the two. The diagram is showing a case where
319 * kernel_total_size is larger than output_size, but this case is
320 * handled by bumping output_size.)
321 *
322 * The above conditions can be illustrated by a diagram:
323 *
324 * 0 output input input+input_size output+init_size
325 * | | | | |
326 * | | | | |
327 * |-----|--------|--------|--------------|-----------|--|-------------|
328 * | | |
329 * | | |
330 * output+init_size-ZO_INIT_SIZE output+output_size output+kernel_total_size
331 *
332 * [output, output+init_size) is the entire memory range used for
333 * extracting the compressed image.
334 *
335 * [output, output+kernel_total_size) is the range needed for the
336 * uncompressed kernel (VO) and its run size (bss, brk, etc).
337 *
338 * [output, output+output_size) is VO plus relocs (i.e. the entire
339 * uncompressed payload contained by ZO). This is the area of the buffer
340 * written to during decompression.
341 *
342 * [output+init_size-ZO_INIT_SIZE, output+init_size) is the worst-case
343 * range of the copied ZO and decompression code. (i.e. the range
344 * covered backwards of size ZO_INIT_SIZE, starting from output+init_size.)
345 *
346 * [input, input+input_size) is the original copied compressed image (ZO)
347 * (i.e. it does not include its run size). This range must be avoided
348 * because it contains the data used for decompression.
349 *
350 * [input+input_size, output+init_size) is [_text, _end) for ZO. This
351 * range includes ZO's heap and stack, and must be avoided since it
352 * performs the decompression.
353 *
354 * Since the above two ranges need to be avoided and they are adjacent,
355 * they can be merged, resulting in: [input, output+init_size) which
356 * becomes the MEM_AVOID_ZO_RANGE below.
357 */
358static void mem_avoid_init(unsigned long input, unsigned long input_size,
359 unsigned long output)
360{
361 unsigned long init_size = boot_params_ptr->hdr.init_size;
362 u64 initrd_start, initrd_size;
363 unsigned long cmd_line, cmd_line_size;
364
365 /*
366 * Avoid the region that is unsafe to overlap during
367 * decompression.
368 */
369 mem_avoid[MEM_AVOID_ZO_RANGE].start = input;
370 mem_avoid[MEM_AVOID_ZO_RANGE].size = (output + init_size) - input;
371
372 /* Avoid initrd. */
373 initrd_start = (u64)boot_params_ptr->ext_ramdisk_image << 32;
374 initrd_start |= boot_params_ptr->hdr.ramdisk_image;
375 initrd_size = (u64)boot_params_ptr->ext_ramdisk_size << 32;
376 initrd_size |= boot_params_ptr->hdr.ramdisk_size;
377 mem_avoid[MEM_AVOID_INITRD].start = initrd_start;
378 mem_avoid[MEM_AVOID_INITRD].size = initrd_size;
379 /* No need to set mapping for initrd, it will be handled in VO. */
380
381 /* Avoid kernel command line. */
382 cmd_line = get_cmd_line_ptr();
383 /* Calculate size of cmd_line. */
384 if (cmd_line) {
385 cmd_line_size = strnlen((char *)cmd_line, COMMAND_LINE_SIZE-1) + 1;
386 mem_avoid[MEM_AVOID_CMDLINE].start = cmd_line;
387 mem_avoid[MEM_AVOID_CMDLINE].size = cmd_line_size;
388 }
389
390 /* Avoid boot parameters. */
391 mem_avoid[MEM_AVOID_BOOTPARAMS].start = (unsigned long)boot_params_ptr;
392 mem_avoid[MEM_AVOID_BOOTPARAMS].size = sizeof(*boot_params_ptr);
393
394 /* We don't need to set a mapping for setup_data. */
395
396 /* Mark the memmap regions we need to avoid */
397 handle_mem_options();
398
399 /* Enumerate the immovable memory regions */
400 num_immovable_mem = count_immovable_mem_regions();
401}
402
403/*
404 * Does this memory vector overlap a known avoided area? If so, record the
405 * overlap region with the lowest address.
406 */
407static bool mem_avoid_overlap(struct mem_vector *img,
408 struct mem_vector *overlap)
409{
410 int i;
411 struct setup_data *ptr;
412 u64 earliest = img->start + img->size;
413 bool is_overlapping = false;
414
415 for (i = 0; i < MEM_AVOID_MAX; i++) {
416 if (mem_overlaps(img, &mem_avoid[i]) &&
417 mem_avoid[i].start < earliest) {
418 *overlap = mem_avoid[i];
419 earliest = overlap->start;
420 is_overlapping = true;
421 }
422 }
423
424 /* Avoid all entries in the setup_data linked list. */
425 ptr = (struct setup_data *)(unsigned long)boot_params_ptr->hdr.setup_data;
426 while (ptr) {
427 struct mem_vector avoid;
428
429 avoid.start = (unsigned long)ptr;
430 avoid.size = sizeof(*ptr) + ptr->len;
431
432 if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) {
433 *overlap = avoid;
434 earliest = overlap->start;
435 is_overlapping = true;
436 }
437
438 if (ptr->type == SETUP_INDIRECT &&
439 ((struct setup_indirect *)ptr->data)->type != SETUP_INDIRECT) {
440 avoid.start = ((struct setup_indirect *)ptr->data)->addr;
441 avoid.size = ((struct setup_indirect *)ptr->data)->len;
442
443 if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) {
444 *overlap = avoid;
445 earliest = overlap->start;
446 is_overlapping = true;
447 }
448 }
449
450 ptr = (struct setup_data *)(unsigned long)ptr->next;
451 }
452
453 return is_overlapping;
454}
455
456struct slot_area {
457 u64 addr;
458 unsigned long num;
459};
460
461#define MAX_SLOT_AREA 100
462
463static struct slot_area slot_areas[MAX_SLOT_AREA];
464static unsigned int slot_area_index;
465static unsigned long slot_max;
466
467static void store_slot_info(struct mem_vector *region, unsigned long image_size)
468{
469 struct slot_area slot_area;
470
471 if (slot_area_index == MAX_SLOT_AREA)
472 return;
473
474 slot_area.addr = region->start;
475 slot_area.num = 1 + (region->size - image_size) / CONFIG_PHYSICAL_ALIGN;
476
477 slot_areas[slot_area_index++] = slot_area;
478 slot_max += slot_area.num;
479}
480
481/*
482 * Skip as many 1GB huge pages as possible in the passed region
483 * according to the number which users specified:
484 */
485static void
486process_gb_huge_pages(struct mem_vector *region, unsigned long image_size)
487{
488 u64 pud_start, pud_end;
489 unsigned long gb_huge_pages;
490 struct mem_vector tmp;
491
492 if (!IS_ENABLED(CONFIG_X86_64) || !max_gb_huge_pages) {
493 store_slot_info(region, image_size);
494 return;
495 }
496
497 /* Are there any 1GB pages in the region? */
498 pud_start = ALIGN(region->start, PUD_SIZE);
499 pud_end = ALIGN_DOWN(region->start + region->size, PUD_SIZE);
500
501 /* No good 1GB huge pages found: */
502 if (pud_start >= pud_end) {
503 store_slot_info(region, image_size);
504 return;
505 }
506
507 /* Check if the head part of the region is usable. */
508 if (pud_start >= region->start + image_size) {
509 tmp.start = region->start;
510 tmp.size = pud_start - region->start;
511 store_slot_info(&tmp, image_size);
512 }
513
514 /* Skip the good 1GB pages. */
515 gb_huge_pages = (pud_end - pud_start) >> PUD_SHIFT;
516 if (gb_huge_pages > max_gb_huge_pages) {
517 pud_end = pud_start + (max_gb_huge_pages << PUD_SHIFT);
518 max_gb_huge_pages = 0;
519 } else {
520 max_gb_huge_pages -= gb_huge_pages;
521 }
522
523 /* Check if the tail part of the region is usable. */
524 if (region->start + region->size >= pud_end + image_size) {
525 tmp.start = pud_end;
526 tmp.size = region->start + region->size - pud_end;
527 store_slot_info(&tmp, image_size);
528 }
529}
530
531static u64 slots_fetch_random(void)
532{
533 unsigned long slot;
534 unsigned int i;
535
536 /* Handle case of no slots stored. */
537 if (slot_max == 0)
538 return 0;
539
540 slot = kaslr_get_random_long("Physical") % slot_max;
541
542 for (i = 0; i < slot_area_index; i++) {
543 if (slot >= slot_areas[i].num) {
544 slot -= slot_areas[i].num;
545 continue;
546 }
547 return slot_areas[i].addr + ((u64)slot * CONFIG_PHYSICAL_ALIGN);
548 }
549
550 if (i == slot_area_index)
551 debug_putstr("slots_fetch_random() failed!?\n");
552 return 0;
553}
554
555static void __process_mem_region(struct mem_vector *entry,
556 unsigned long minimum,
557 unsigned long image_size)
558{
559 struct mem_vector region, overlap;
560 u64 region_end;
561
562 /* Enforce minimum and memory limit. */
563 region.start = max_t(u64, entry->start, minimum);
564 region_end = min(entry->start + entry->size, mem_limit);
565
566 /* Give up if slot area array is full. */
567 while (slot_area_index < MAX_SLOT_AREA) {
568 /* Potentially raise address to meet alignment needs. */
569 region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
570
571 /* Did we raise the address above the passed in memory entry? */
572 if (region.start > region_end)
573 return;
574
575 /* Reduce size by any delta from the original address. */
576 region.size = region_end - region.start;
577
578 /* Return if region can't contain decompressed kernel */
579 if (region.size < image_size)
580 return;
581
582 /* If nothing overlaps, store the region and return. */
583 if (!mem_avoid_overlap(®ion, &overlap)) {
584 process_gb_huge_pages(®ion, image_size);
585 return;
586 }
587
588 /* Store beginning of region if holds at least image_size. */
589 if (overlap.start >= region.start + image_size) {
590 region.size = overlap.start - region.start;
591 process_gb_huge_pages(®ion, image_size);
592 }
593
594 /* Clip off the overlapping region and start over. */
595 region.start = overlap.start + overlap.size;
596 }
597}
598
599static bool process_mem_region(struct mem_vector *region,
600 unsigned long minimum,
601 unsigned long image_size)
602{
603 int i;
604 /*
605 * If no immovable memory found, or MEMORY_HOTREMOVE disabled,
606 * use @region directly.
607 */
608 if (!num_immovable_mem) {
609 __process_mem_region(region, minimum, image_size);
610
611 if (slot_area_index == MAX_SLOT_AREA) {
612 debug_putstr("Aborted e820/efi memmap scan (slot_areas full)!\n");
613 return true;
614 }
615 return false;
616 }
617
618#if defined(CONFIG_MEMORY_HOTREMOVE) && defined(CONFIG_ACPI)
619 /*
620 * If immovable memory found, filter the intersection between
621 * immovable memory and @region.
622 */
623 for (i = 0; i < num_immovable_mem; i++) {
624 u64 start, end, entry_end, region_end;
625 struct mem_vector entry;
626
627 if (!mem_overlaps(region, &immovable_mem[i]))
628 continue;
629
630 start = immovable_mem[i].start;
631 end = start + immovable_mem[i].size;
632 region_end = region->start + region->size;
633
634 entry.start = clamp(region->start, start, end);
635 entry_end = clamp(region_end, start, end);
636 entry.size = entry_end - entry.start;
637
638 __process_mem_region(&entry, minimum, image_size);
639
640 if (slot_area_index == MAX_SLOT_AREA) {
641 debug_putstr("Aborted e820/efi memmap scan when walking immovable regions(slot_areas full)!\n");
642 return true;
643 }
644 }
645#endif
646 return false;
647}
648
649#ifdef CONFIG_EFI
650
651/*
652 * Only EFI_CONVENTIONAL_MEMORY and EFI_UNACCEPTED_MEMORY (if supported) are
653 * guaranteed to be free.
654 *
655 * Pick free memory more conservatively than the EFI spec allows: according to
656 * the spec, EFI_BOOT_SERVICES_{CODE|DATA} are also free memory and thus
657 * available to place the kernel image into, but in practice there's firmware
658 * where using that memory leads to crashes. Buggy vendor EFI code registers
659 * for an event that triggers on SetVirtualAddressMap(). The handler assumes
660 * that EFI_BOOT_SERVICES_DATA memory has not been touched by loader yet, which
661 * is probably true for Windows.
662 *
663 * Preserve EFI_BOOT_SERVICES_* regions until after SetVirtualAddressMap().
664 */
665static inline bool memory_type_is_free(efi_memory_desc_t *md)
666{
667 if (md->type == EFI_CONVENTIONAL_MEMORY)
668 return true;
669
670 if (IS_ENABLED(CONFIG_UNACCEPTED_MEMORY) &&
671 md->type == EFI_UNACCEPTED_MEMORY)
672 return true;
673
674 return false;
675}
676
677/*
678 * Returns true if we processed the EFI memmap, which we prefer over the E820
679 * table if it is available.
680 */
681static bool
682process_efi_entries(unsigned long minimum, unsigned long image_size)
683{
684 struct efi_info *e = &boot_params_ptr->efi_info;
685 bool efi_mirror_found = false;
686 struct mem_vector region;
687 efi_memory_desc_t *md;
688 unsigned long pmap;
689 char *signature;
690 u32 nr_desc;
691 int i;
692
693 signature = (char *)&e->efi_loader_signature;
694 if (strncmp(signature, EFI32_LOADER_SIGNATURE, 4) &&
695 strncmp(signature, EFI64_LOADER_SIGNATURE, 4))
696 return false;
697
698#ifdef CONFIG_X86_32
699 /* Can't handle data above 4GB at this time */
700 if (e->efi_memmap_hi) {
701 warn("EFI memmap is above 4GB, can't be handled now on x86_32. EFI should be disabled.\n");
702 return false;
703 }
704 pmap = e->efi_memmap;
705#else
706 pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
707#endif
708
709 nr_desc = e->efi_memmap_size / e->efi_memdesc_size;
710 for (i = 0; i < nr_desc; i++) {
711 md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
712 if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
713 efi_mirror_found = true;
714 break;
715 }
716 }
717
718 for (i = 0; i < nr_desc; i++) {
719 md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
720
721 if (!memory_type_is_free(md))
722 continue;
723
724 if (efi_soft_reserve_enabled() &&
725 (md->attribute & EFI_MEMORY_SP))
726 continue;
727
728 if (efi_mirror_found &&
729 !(md->attribute & EFI_MEMORY_MORE_RELIABLE))
730 continue;
731
732 region.start = md->phys_addr;
733 region.size = md->num_pages << EFI_PAGE_SHIFT;
734 if (process_mem_region(®ion, minimum, image_size))
735 break;
736 }
737 return true;
738}
739#else
740static inline bool
741process_efi_entries(unsigned long minimum, unsigned long image_size)
742{
743 return false;
744}
745#endif
746
747static void process_e820_entries(unsigned long minimum,
748 unsigned long image_size)
749{
750 int i;
751 struct mem_vector region;
752 struct boot_e820_entry *entry;
753
754 /* Verify potential e820 positions, appending to slots list. */
755 for (i = 0; i < boot_params_ptr->e820_entries; i++) {
756 entry = &boot_params_ptr->e820_table[i];
757 /* Skip non-RAM entries. */
758 if (entry->type != E820_TYPE_RAM)
759 continue;
760 region.start = entry->addr;
761 region.size = entry->size;
762 if (process_mem_region(®ion, minimum, image_size))
763 break;
764 }
765}
766
767static unsigned long find_random_phys_addr(unsigned long minimum,
768 unsigned long image_size)
769{
770 u64 phys_addr;
771
772 /* Bail out early if it's impossible to succeed. */
773 if (minimum + image_size > mem_limit)
774 return 0;
775
776 /* Check if we had too many memmaps. */
777 if (memmap_too_large) {
778 debug_putstr("Aborted memory entries scan (more than 4 memmap= args)!\n");
779 return 0;
780 }
781
782 if (!process_efi_entries(minimum, image_size))
783 process_e820_entries(minimum, image_size);
784
785 phys_addr = slots_fetch_random();
786
787 /* Perform a final check to make sure the address is in range. */
788 if (phys_addr < minimum || phys_addr + image_size > mem_limit) {
789 warn("Invalid physical address chosen!\n");
790 return 0;
791 }
792
793 return (unsigned long)phys_addr;
794}
795
796static unsigned long find_random_virt_addr(unsigned long minimum,
797 unsigned long image_size)
798{
799 unsigned long slots, random_addr;
800
801 /*
802 * There are how many CONFIG_PHYSICAL_ALIGN-sized slots
803 * that can hold image_size within the range of minimum to
804 * KERNEL_IMAGE_SIZE?
805 */
806 slots = 1 + (KERNEL_IMAGE_SIZE - minimum - image_size) / CONFIG_PHYSICAL_ALIGN;
807
808 random_addr = kaslr_get_random_long("Virtual") % slots;
809
810 return random_addr * CONFIG_PHYSICAL_ALIGN + minimum;
811}
812
813/*
814 * Since this function examines addresses much more numerically,
815 * it takes the input and output pointers as 'unsigned long'.
816 */
817void choose_random_location(unsigned long input,
818 unsigned long input_size,
819 unsigned long *output,
820 unsigned long output_size,
821 unsigned long *virt_addr)
822{
823 unsigned long random_addr, min_addr;
824
825 if (cmdline_find_option_bool("nokaslr")) {
826 warn("KASLR disabled: 'nokaslr' on cmdline.");
827 return;
828 }
829
830 boot_params_ptr->hdr.loadflags |= KASLR_FLAG;
831
832 if (IS_ENABLED(CONFIG_X86_32))
833 mem_limit = KERNEL_IMAGE_SIZE;
834 else
835 mem_limit = MAXMEM;
836
837 /* Record the various known unsafe memory ranges. */
838 mem_avoid_init(input, input_size, *output);
839
840 /*
841 * Low end of the randomization range should be the
842 * smaller of 512M or the initial kernel image
843 * location:
844 */
845 min_addr = min(*output, 512UL << 20);
846 /* Make sure minimum is aligned. */
847 min_addr = ALIGN(min_addr, CONFIG_PHYSICAL_ALIGN);
848
849 /* Walk available memory entries to find a random address. */
850 random_addr = find_random_phys_addr(min_addr, output_size);
851 if (!random_addr) {
852 warn("Physical KASLR disabled: no suitable memory region!");
853 } else {
854 /* Update the new physical address location. */
855 if (*output != random_addr)
856 *output = random_addr;
857 }
858
859
860 /* Pick random virtual address starting from LOAD_PHYSICAL_ADDR. */
861 if (IS_ENABLED(CONFIG_X86_64))
862 random_addr = find_random_virt_addr(LOAD_PHYSICAL_ADDR, output_size);
863 *virt_addr = random_addr;
864}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * kaslr.c
4 *
5 * This contains the routines needed to generate a reasonable level of
6 * entropy to choose a randomized kernel base address offset in support
7 * of Kernel Address Space Layout Randomization (KASLR). Additionally
8 * handles walking the physical memory maps (and tracking memory regions
9 * to avoid) in order to select a physical memory location that can
10 * contain the entire properly aligned running kernel image.
11 *
12 */
13
14/*
15 * isspace() in linux/ctype.h is expected by next_args() to filter
16 * out "space/lf/tab". While boot/ctype.h conflicts with linux/ctype.h,
17 * since isdigit() is implemented in both of them. Hence disable it
18 * here.
19 */
20#define BOOT_CTYPE_H
21
22#include "misc.h"
23#include "error.h"
24#include "../string.h"
25
26#include <generated/compile.h>
27#include <linux/module.h>
28#include <linux/uts.h>
29#include <linux/utsname.h>
30#include <linux/ctype.h>
31#include <linux/efi.h>
32#include <generated/utsrelease.h>
33#include <asm/efi.h>
34
35/* Macros used by the included decompressor code below. */
36#define STATIC
37#include <linux/decompress/mm.h>
38
39#ifdef CONFIG_X86_5LEVEL
40unsigned int __pgtable_l5_enabled;
41unsigned int pgdir_shift __ro_after_init = 39;
42unsigned int ptrs_per_p4d __ro_after_init = 1;
43#endif
44
45extern unsigned long get_cmd_line_ptr(void);
46
47/* Used by PAGE_KERN* macros: */
48pteval_t __default_kernel_pte_mask __read_mostly = ~0;
49
50/* Simplified build-specific string for starting entropy. */
51static const char build_str[] = UTS_RELEASE " (" LINUX_COMPILE_BY "@"
52 LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION;
53
54static unsigned long rotate_xor(unsigned long hash, const void *area,
55 size_t size)
56{
57 size_t i;
58 unsigned long *ptr = (unsigned long *)area;
59
60 for (i = 0; i < size / sizeof(hash); i++) {
61 /* Rotate by odd number of bits and XOR. */
62 hash = (hash << ((sizeof(hash) * 8) - 7)) | (hash >> 7);
63 hash ^= ptr[i];
64 }
65
66 return hash;
67}
68
69/* Attempt to create a simple but unpredictable starting entropy. */
70static unsigned long get_boot_seed(void)
71{
72 unsigned long hash = 0;
73
74 hash = rotate_xor(hash, build_str, sizeof(build_str));
75 hash = rotate_xor(hash, boot_params, sizeof(*boot_params));
76
77 return hash;
78}
79
80#define KASLR_COMPRESSED_BOOT
81#include "../../lib/kaslr.c"
82
83
84/* Only supporting at most 4 unusable memmap regions with kaslr */
85#define MAX_MEMMAP_REGIONS 4
86
87static bool memmap_too_large;
88
89
90/* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */
91static unsigned long long mem_limit = ULLONG_MAX;
92
93/* Number of immovable memory regions */
94static int num_immovable_mem;
95
96enum mem_avoid_index {
97 MEM_AVOID_ZO_RANGE = 0,
98 MEM_AVOID_INITRD,
99 MEM_AVOID_CMDLINE,
100 MEM_AVOID_BOOTPARAMS,
101 MEM_AVOID_MEMMAP_BEGIN,
102 MEM_AVOID_MEMMAP_END = MEM_AVOID_MEMMAP_BEGIN + MAX_MEMMAP_REGIONS - 1,
103 MEM_AVOID_MAX,
104};
105
106static struct mem_vector mem_avoid[MEM_AVOID_MAX];
107
108static bool mem_overlaps(struct mem_vector *one, struct mem_vector *two)
109{
110 /* Item one is entirely before item two. */
111 if (one->start + one->size <= two->start)
112 return false;
113 /* Item one is entirely after item two. */
114 if (one->start >= two->start + two->size)
115 return false;
116 return true;
117}
118
119char *skip_spaces(const char *str)
120{
121 while (isspace(*str))
122 ++str;
123 return (char *)str;
124}
125#include "../../../../lib/ctype.c"
126#include "../../../../lib/cmdline.c"
127
128enum parse_mode {
129 PARSE_MEMMAP,
130 PARSE_EFI,
131};
132
133static int
134parse_memmap(char *p, unsigned long long *start, unsigned long long *size,
135 enum parse_mode mode)
136{
137 char *oldp;
138
139 if (!p)
140 return -EINVAL;
141
142 /* We don't care about this option here */
143 if (!strncmp(p, "exactmap", 8))
144 return -EINVAL;
145
146 oldp = p;
147 *size = memparse(p, &p);
148 if (p == oldp)
149 return -EINVAL;
150
151 switch (*p) {
152 case '#':
153 case '$':
154 case '!':
155 *start = memparse(p + 1, &p);
156 return 0;
157 case '@':
158 if (mode == PARSE_MEMMAP) {
159 /*
160 * memmap=nn@ss specifies usable region, should
161 * be skipped
162 */
163 *size = 0;
164 } else {
165 unsigned long long flags;
166
167 /*
168 * efi_fake_mem=nn@ss:attr the attr specifies
169 * flags that might imply a soft-reservation.
170 */
171 *start = memparse(p + 1, &p);
172 if (p && *p == ':') {
173 p++;
174 if (kstrtoull(p, 0, &flags) < 0)
175 *size = 0;
176 else if (flags & EFI_MEMORY_SP)
177 return 0;
178 }
179 *size = 0;
180 }
181 fallthrough;
182 default:
183 /*
184 * If w/o offset, only size specified, memmap=nn[KMG] has the
185 * same behaviour as mem=nn[KMG]. It limits the max address
186 * system can use. Region above the limit should be avoided.
187 */
188 *start = 0;
189 return 0;
190 }
191
192 return -EINVAL;
193}
194
195static void mem_avoid_memmap(enum parse_mode mode, char *str)
196{
197 static int i;
198
199 if (i >= MAX_MEMMAP_REGIONS)
200 return;
201
202 while (str && (i < MAX_MEMMAP_REGIONS)) {
203 int rc;
204 unsigned long long start, size;
205 char *k = strchr(str, ',');
206
207 if (k)
208 *k++ = 0;
209
210 rc = parse_memmap(str, &start, &size, mode);
211 if (rc < 0)
212 break;
213 str = k;
214
215 if (start == 0) {
216 /* Store the specified memory limit if size > 0 */
217 if (size > 0)
218 mem_limit = size;
219
220 continue;
221 }
222
223 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
224 mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
225 i++;
226 }
227
228 /* More than 4 memmaps, fail kaslr */
229 if ((i >= MAX_MEMMAP_REGIONS) && str)
230 memmap_too_large = true;
231}
232
233/* Store the number of 1GB huge pages which users specified: */
234static unsigned long max_gb_huge_pages;
235
236static void parse_gb_huge_pages(char *param, char *val)
237{
238 static bool gbpage_sz;
239 char *p;
240
241 if (!strcmp(param, "hugepagesz")) {
242 p = val;
243 if (memparse(p, &p) != PUD_SIZE) {
244 gbpage_sz = false;
245 return;
246 }
247
248 if (gbpage_sz)
249 warn("Repeatedly set hugeTLB page size of 1G!\n");
250 gbpage_sz = true;
251 return;
252 }
253
254 if (!strcmp(param, "hugepages") && gbpage_sz) {
255 p = val;
256 max_gb_huge_pages = simple_strtoull(p, &p, 0);
257 return;
258 }
259}
260
261static void handle_mem_options(void)
262{
263 char *args = (char *)get_cmd_line_ptr();
264 size_t len = strlen((char *)args);
265 char *tmp_cmdline;
266 char *param, *val;
267 u64 mem_size;
268
269 if (!strstr(args, "memmap=") && !strstr(args, "mem=") &&
270 !strstr(args, "hugepages"))
271 return;
272
273 tmp_cmdline = malloc(len + 1);
274 if (!tmp_cmdline)
275 error("Failed to allocate space for tmp_cmdline");
276
277 memcpy(tmp_cmdline, args, len);
278 tmp_cmdline[len] = 0;
279 args = tmp_cmdline;
280
281 /* Chew leading spaces */
282 args = skip_spaces(args);
283
284 while (*args) {
285 args = next_arg(args, ¶m, &val);
286 /* Stop at -- */
287 if (!val && strcmp(param, "--") == 0) {
288 warn("Only '--' specified in cmdline");
289 goto out;
290 }
291
292 if (!strcmp(param, "memmap")) {
293 mem_avoid_memmap(PARSE_MEMMAP, val);
294 } else if (strstr(param, "hugepages")) {
295 parse_gb_huge_pages(param, val);
296 } else if (!strcmp(param, "mem")) {
297 char *p = val;
298
299 if (!strcmp(p, "nopentium"))
300 continue;
301 mem_size = memparse(p, &p);
302 if (mem_size == 0)
303 goto out;
304
305 mem_limit = mem_size;
306 } else if (!strcmp(param, "efi_fake_mem")) {
307 mem_avoid_memmap(PARSE_EFI, val);
308 }
309 }
310
311out:
312 free(tmp_cmdline);
313 return;
314}
315
316/*
317 * In theory, KASLR can put the kernel anywhere in the range of [16M, 64T).
318 * The mem_avoid array is used to store the ranges that need to be avoided
319 * when KASLR searches for an appropriate random address. We must avoid any
320 * regions that are unsafe to overlap with during decompression, and other
321 * things like the initrd, cmdline and boot_params. This comment seeks to
322 * explain mem_avoid as clearly as possible since incorrect mem_avoid
323 * memory ranges lead to really hard to debug boot failures.
324 *
325 * The initrd, cmdline, and boot_params are trivial to identify for
326 * avoiding. They are MEM_AVOID_INITRD, MEM_AVOID_CMDLINE, and
327 * MEM_AVOID_BOOTPARAMS respectively below.
328 *
329 * What is not obvious how to avoid is the range of memory that is used
330 * during decompression (MEM_AVOID_ZO_RANGE below). This range must cover
331 * the compressed kernel (ZO) and its run space, which is used to extract
332 * the uncompressed kernel (VO) and relocs.
333 *
334 * ZO's full run size sits against the end of the decompression buffer, so
335 * we can calculate where text, data, bss, etc of ZO are positioned more
336 * easily.
337 *
338 * For additional background, the decompression calculations can be found
339 * in header.S, and the memory diagram is based on the one found in misc.c.
340 *
341 * The following conditions are already enforced by the image layouts and
342 * associated code:
343 * - input + input_size >= output + output_size
344 * - kernel_total_size <= init_size
345 * - kernel_total_size <= output_size (see Note below)
346 * - output + init_size >= output + output_size
347 *
348 * (Note that kernel_total_size and output_size have no fundamental
349 * relationship, but output_size is passed to choose_random_location
350 * as a maximum of the two. The diagram is showing a case where
351 * kernel_total_size is larger than output_size, but this case is
352 * handled by bumping output_size.)
353 *
354 * The above conditions can be illustrated by a diagram:
355 *
356 * 0 output input input+input_size output+init_size
357 * | | | | |
358 * | | | | |
359 * |-----|--------|--------|--------------|-----------|--|-------------|
360 * | | |
361 * | | |
362 * output+init_size-ZO_INIT_SIZE output+output_size output+kernel_total_size
363 *
364 * [output, output+init_size) is the entire memory range used for
365 * extracting the compressed image.
366 *
367 * [output, output+kernel_total_size) is the range needed for the
368 * uncompressed kernel (VO) and its run size (bss, brk, etc).
369 *
370 * [output, output+output_size) is VO plus relocs (i.e. the entire
371 * uncompressed payload contained by ZO). This is the area of the buffer
372 * written to during decompression.
373 *
374 * [output+init_size-ZO_INIT_SIZE, output+init_size) is the worst-case
375 * range of the copied ZO and decompression code. (i.e. the range
376 * covered backwards of size ZO_INIT_SIZE, starting from output+init_size.)
377 *
378 * [input, input+input_size) is the original copied compressed image (ZO)
379 * (i.e. it does not include its run size). This range must be avoided
380 * because it contains the data used for decompression.
381 *
382 * [input+input_size, output+init_size) is [_text, _end) for ZO. This
383 * range includes ZO's heap and stack, and must be avoided since it
384 * performs the decompression.
385 *
386 * Since the above two ranges need to be avoided and they are adjacent,
387 * they can be merged, resulting in: [input, output+init_size) which
388 * becomes the MEM_AVOID_ZO_RANGE below.
389 */
390static void mem_avoid_init(unsigned long input, unsigned long input_size,
391 unsigned long output)
392{
393 unsigned long init_size = boot_params->hdr.init_size;
394 u64 initrd_start, initrd_size;
395 u64 cmd_line, cmd_line_size;
396 char *ptr;
397
398 /*
399 * Avoid the region that is unsafe to overlap during
400 * decompression.
401 */
402 mem_avoid[MEM_AVOID_ZO_RANGE].start = input;
403 mem_avoid[MEM_AVOID_ZO_RANGE].size = (output + init_size) - input;
404 add_identity_map(mem_avoid[MEM_AVOID_ZO_RANGE].start,
405 mem_avoid[MEM_AVOID_ZO_RANGE].size);
406
407 /* Avoid initrd. */
408 initrd_start = (u64)boot_params->ext_ramdisk_image << 32;
409 initrd_start |= boot_params->hdr.ramdisk_image;
410 initrd_size = (u64)boot_params->ext_ramdisk_size << 32;
411 initrd_size |= boot_params->hdr.ramdisk_size;
412 mem_avoid[MEM_AVOID_INITRD].start = initrd_start;
413 mem_avoid[MEM_AVOID_INITRD].size = initrd_size;
414 /* No need to set mapping for initrd, it will be handled in VO. */
415
416 /* Avoid kernel command line. */
417 cmd_line = (u64)boot_params->ext_cmd_line_ptr << 32;
418 cmd_line |= boot_params->hdr.cmd_line_ptr;
419 /* Calculate size of cmd_line. */
420 ptr = (char *)(unsigned long)cmd_line;
421 for (cmd_line_size = 0; ptr[cmd_line_size++];)
422 ;
423 mem_avoid[MEM_AVOID_CMDLINE].start = cmd_line;
424 mem_avoid[MEM_AVOID_CMDLINE].size = cmd_line_size;
425 add_identity_map(mem_avoid[MEM_AVOID_CMDLINE].start,
426 mem_avoid[MEM_AVOID_CMDLINE].size);
427
428 /* Avoid boot parameters. */
429 mem_avoid[MEM_AVOID_BOOTPARAMS].start = (unsigned long)boot_params;
430 mem_avoid[MEM_AVOID_BOOTPARAMS].size = sizeof(*boot_params);
431 add_identity_map(mem_avoid[MEM_AVOID_BOOTPARAMS].start,
432 mem_avoid[MEM_AVOID_BOOTPARAMS].size);
433
434 /* We don't need to set a mapping for setup_data. */
435
436 /* Mark the memmap regions we need to avoid */
437 handle_mem_options();
438
439 /* Enumerate the immovable memory regions */
440 num_immovable_mem = count_immovable_mem_regions();
441
442#ifdef CONFIG_X86_VERBOSE_BOOTUP
443 /* Make sure video RAM can be used. */
444 add_identity_map(0, PMD_SIZE);
445#endif
446}
447
448/*
449 * Does this memory vector overlap a known avoided area? If so, record the
450 * overlap region with the lowest address.
451 */
452static bool mem_avoid_overlap(struct mem_vector *img,
453 struct mem_vector *overlap)
454{
455 int i;
456 struct setup_data *ptr;
457 unsigned long earliest = img->start + img->size;
458 bool is_overlapping = false;
459
460 for (i = 0; i < MEM_AVOID_MAX; i++) {
461 if (mem_overlaps(img, &mem_avoid[i]) &&
462 mem_avoid[i].start < earliest) {
463 *overlap = mem_avoid[i];
464 earliest = overlap->start;
465 is_overlapping = true;
466 }
467 }
468
469 /* Avoid all entries in the setup_data linked list. */
470 ptr = (struct setup_data *)(unsigned long)boot_params->hdr.setup_data;
471 while (ptr) {
472 struct mem_vector avoid;
473
474 avoid.start = (unsigned long)ptr;
475 avoid.size = sizeof(*ptr) + ptr->len;
476
477 if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) {
478 *overlap = avoid;
479 earliest = overlap->start;
480 is_overlapping = true;
481 }
482
483 if (ptr->type == SETUP_INDIRECT &&
484 ((struct setup_indirect *)ptr->data)->type != SETUP_INDIRECT) {
485 avoid.start = ((struct setup_indirect *)ptr->data)->addr;
486 avoid.size = ((struct setup_indirect *)ptr->data)->len;
487
488 if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) {
489 *overlap = avoid;
490 earliest = overlap->start;
491 is_overlapping = true;
492 }
493 }
494
495 ptr = (struct setup_data *)(unsigned long)ptr->next;
496 }
497
498 return is_overlapping;
499}
500
501struct slot_area {
502 unsigned long addr;
503 int num;
504};
505
506#define MAX_SLOT_AREA 100
507
508static struct slot_area slot_areas[MAX_SLOT_AREA];
509
510static unsigned long slot_max;
511
512static unsigned long slot_area_index;
513
514static void store_slot_info(struct mem_vector *region, unsigned long image_size)
515{
516 struct slot_area slot_area;
517
518 if (slot_area_index == MAX_SLOT_AREA)
519 return;
520
521 slot_area.addr = region->start;
522 slot_area.num = (region->size - image_size) /
523 CONFIG_PHYSICAL_ALIGN + 1;
524
525 if (slot_area.num > 0) {
526 slot_areas[slot_area_index++] = slot_area;
527 slot_max += slot_area.num;
528 }
529}
530
531/*
532 * Skip as many 1GB huge pages as possible in the passed region
533 * according to the number which users specified:
534 */
535static void
536process_gb_huge_pages(struct mem_vector *region, unsigned long image_size)
537{
538 unsigned long addr, size = 0;
539 struct mem_vector tmp;
540 int i = 0;
541
542 if (!max_gb_huge_pages) {
543 store_slot_info(region, image_size);
544 return;
545 }
546
547 addr = ALIGN(region->start, PUD_SIZE);
548 /* Did we raise the address above the passed in memory entry? */
549 if (addr < region->start + region->size)
550 size = region->size - (addr - region->start);
551
552 /* Check how many 1GB huge pages can be filtered out: */
553 while (size > PUD_SIZE && max_gb_huge_pages) {
554 size -= PUD_SIZE;
555 max_gb_huge_pages--;
556 i++;
557 }
558
559 /* No good 1GB huge pages found: */
560 if (!i) {
561 store_slot_info(region, image_size);
562 return;
563 }
564
565 /*
566 * Skip those 'i'*1GB good huge pages, and continue checking and
567 * processing the remaining head or tail part of the passed region
568 * if available.
569 */
570
571 if (addr >= region->start + image_size) {
572 tmp.start = region->start;
573 tmp.size = addr - region->start;
574 store_slot_info(&tmp, image_size);
575 }
576
577 size = region->size - (addr - region->start) - i * PUD_SIZE;
578 if (size >= image_size) {
579 tmp.start = addr + i * PUD_SIZE;
580 tmp.size = size;
581 store_slot_info(&tmp, image_size);
582 }
583}
584
585static unsigned long slots_fetch_random(void)
586{
587 unsigned long slot;
588 int i;
589
590 /* Handle case of no slots stored. */
591 if (slot_max == 0)
592 return 0;
593
594 slot = kaslr_get_random_long("Physical") % slot_max;
595
596 for (i = 0; i < slot_area_index; i++) {
597 if (slot >= slot_areas[i].num) {
598 slot -= slot_areas[i].num;
599 continue;
600 }
601 return slot_areas[i].addr + slot * CONFIG_PHYSICAL_ALIGN;
602 }
603
604 if (i == slot_area_index)
605 debug_putstr("slots_fetch_random() failed!?\n");
606 return 0;
607}
608
609static void __process_mem_region(struct mem_vector *entry,
610 unsigned long minimum,
611 unsigned long image_size)
612{
613 struct mem_vector region, overlap;
614 unsigned long start_orig, end;
615 struct mem_vector cur_entry;
616
617 /* On 32-bit, ignore entries entirely above our maximum. */
618 if (IS_ENABLED(CONFIG_X86_32) && entry->start >= KERNEL_IMAGE_SIZE)
619 return;
620
621 /* Ignore entries entirely below our minimum. */
622 if (entry->start + entry->size < minimum)
623 return;
624
625 /* Ignore entries above memory limit */
626 end = min(entry->size + entry->start, mem_limit);
627 if (entry->start >= end)
628 return;
629 cur_entry.start = entry->start;
630 cur_entry.size = end - entry->start;
631
632 region.start = cur_entry.start;
633 region.size = cur_entry.size;
634
635 /* Give up if slot area array is full. */
636 while (slot_area_index < MAX_SLOT_AREA) {
637 start_orig = region.start;
638
639 /* Potentially raise address to minimum location. */
640 if (region.start < minimum)
641 region.start = minimum;
642
643 /* Potentially raise address to meet alignment needs. */
644 region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
645
646 /* Did we raise the address above the passed in memory entry? */
647 if (region.start > cur_entry.start + cur_entry.size)
648 return;
649
650 /* Reduce size by any delta from the original address. */
651 region.size -= region.start - start_orig;
652
653 /* On 32-bit, reduce region size to fit within max size. */
654 if (IS_ENABLED(CONFIG_X86_32) &&
655 region.start + region.size > KERNEL_IMAGE_SIZE)
656 region.size = KERNEL_IMAGE_SIZE - region.start;
657
658 /* Return if region can't contain decompressed kernel */
659 if (region.size < image_size)
660 return;
661
662 /* If nothing overlaps, store the region and return. */
663 if (!mem_avoid_overlap(®ion, &overlap)) {
664 process_gb_huge_pages(®ion, image_size);
665 return;
666 }
667
668 /* Store beginning of region if holds at least image_size. */
669 if (overlap.start > region.start + image_size) {
670 struct mem_vector beginning;
671
672 beginning.start = region.start;
673 beginning.size = overlap.start - region.start;
674 process_gb_huge_pages(&beginning, image_size);
675 }
676
677 /* Return if overlap extends to or past end of region. */
678 if (overlap.start + overlap.size >= region.start + region.size)
679 return;
680
681 /* Clip off the overlapping region and start over. */
682 region.size -= overlap.start - region.start + overlap.size;
683 region.start = overlap.start + overlap.size;
684 }
685}
686
687static bool process_mem_region(struct mem_vector *region,
688 unsigned long long minimum,
689 unsigned long long image_size)
690{
691 int i;
692 /*
693 * If no immovable memory found, or MEMORY_HOTREMOVE disabled,
694 * use @region directly.
695 */
696 if (!num_immovable_mem) {
697 __process_mem_region(region, minimum, image_size);
698
699 if (slot_area_index == MAX_SLOT_AREA) {
700 debug_putstr("Aborted e820/efi memmap scan (slot_areas full)!\n");
701 return 1;
702 }
703 return 0;
704 }
705
706#if defined(CONFIG_MEMORY_HOTREMOVE) && defined(CONFIG_ACPI)
707 /*
708 * If immovable memory found, filter the intersection between
709 * immovable memory and @region.
710 */
711 for (i = 0; i < num_immovable_mem; i++) {
712 unsigned long long start, end, entry_end, region_end;
713 struct mem_vector entry;
714
715 if (!mem_overlaps(region, &immovable_mem[i]))
716 continue;
717
718 start = immovable_mem[i].start;
719 end = start + immovable_mem[i].size;
720 region_end = region->start + region->size;
721
722 entry.start = clamp(region->start, start, end);
723 entry_end = clamp(region_end, start, end);
724 entry.size = entry_end - entry.start;
725
726 __process_mem_region(&entry, minimum, image_size);
727
728 if (slot_area_index == MAX_SLOT_AREA) {
729 debug_putstr("Aborted e820/efi memmap scan when walking immovable regions(slot_areas full)!\n");
730 return 1;
731 }
732 }
733#endif
734 return 0;
735}
736
737#ifdef CONFIG_EFI
738/*
739 * Returns true if mirror region found (and must have been processed
740 * for slots adding)
741 */
742static bool
743process_efi_entries(unsigned long minimum, unsigned long image_size)
744{
745 struct efi_info *e = &boot_params->efi_info;
746 bool efi_mirror_found = false;
747 struct mem_vector region;
748 efi_memory_desc_t *md;
749 unsigned long pmap;
750 char *signature;
751 u32 nr_desc;
752 int i;
753
754 signature = (char *)&e->efi_loader_signature;
755 if (strncmp(signature, EFI32_LOADER_SIGNATURE, 4) &&
756 strncmp(signature, EFI64_LOADER_SIGNATURE, 4))
757 return false;
758
759#ifdef CONFIG_X86_32
760 /* Can't handle data above 4GB at this time */
761 if (e->efi_memmap_hi) {
762 warn("EFI memmap is above 4GB, can't be handled now on x86_32. EFI should be disabled.\n");
763 return false;
764 }
765 pmap = e->efi_memmap;
766#else
767 pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
768#endif
769
770 nr_desc = e->efi_memmap_size / e->efi_memdesc_size;
771 for (i = 0; i < nr_desc; i++) {
772 md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
773 if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
774 efi_mirror_found = true;
775 break;
776 }
777 }
778
779 for (i = 0; i < nr_desc; i++) {
780 md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
781
782 /*
783 * Here we are more conservative in picking free memory than
784 * the EFI spec allows:
785 *
786 * According to the spec, EFI_BOOT_SERVICES_{CODE|DATA} are also
787 * free memory and thus available to place the kernel image into,
788 * but in practice there's firmware where using that memory leads
789 * to crashes.
790 *
791 * Only EFI_CONVENTIONAL_MEMORY is guaranteed to be free.
792 */
793 if (md->type != EFI_CONVENTIONAL_MEMORY)
794 continue;
795
796 if (efi_soft_reserve_enabled() &&
797 (md->attribute & EFI_MEMORY_SP))
798 continue;
799
800 if (efi_mirror_found &&
801 !(md->attribute & EFI_MEMORY_MORE_RELIABLE))
802 continue;
803
804 region.start = md->phys_addr;
805 region.size = md->num_pages << EFI_PAGE_SHIFT;
806 if (process_mem_region(®ion, minimum, image_size))
807 break;
808 }
809 return true;
810}
811#else
812static inline bool
813process_efi_entries(unsigned long minimum, unsigned long image_size)
814{
815 return false;
816}
817#endif
818
819static void process_e820_entries(unsigned long minimum,
820 unsigned long image_size)
821{
822 int i;
823 struct mem_vector region;
824 struct boot_e820_entry *entry;
825
826 /* Verify potential e820 positions, appending to slots list. */
827 for (i = 0; i < boot_params->e820_entries; i++) {
828 entry = &boot_params->e820_table[i];
829 /* Skip non-RAM entries. */
830 if (entry->type != E820_TYPE_RAM)
831 continue;
832 region.start = entry->addr;
833 region.size = entry->size;
834 if (process_mem_region(®ion, minimum, image_size))
835 break;
836 }
837}
838
839static unsigned long find_random_phys_addr(unsigned long minimum,
840 unsigned long image_size)
841{
842 /* Check if we had too many memmaps. */
843 if (memmap_too_large) {
844 debug_putstr("Aborted memory entries scan (more than 4 memmap= args)!\n");
845 return 0;
846 }
847
848 /* Make sure minimum is aligned. */
849 minimum = ALIGN(minimum, CONFIG_PHYSICAL_ALIGN);
850
851 if (process_efi_entries(minimum, image_size))
852 return slots_fetch_random();
853
854 process_e820_entries(minimum, image_size);
855 return slots_fetch_random();
856}
857
858static unsigned long find_random_virt_addr(unsigned long minimum,
859 unsigned long image_size)
860{
861 unsigned long slots, random_addr;
862
863 /* Make sure minimum is aligned. */
864 minimum = ALIGN(minimum, CONFIG_PHYSICAL_ALIGN);
865 /* Align image_size for easy slot calculations. */
866 image_size = ALIGN(image_size, CONFIG_PHYSICAL_ALIGN);
867
868 /*
869 * There are how many CONFIG_PHYSICAL_ALIGN-sized slots
870 * that can hold image_size within the range of minimum to
871 * KERNEL_IMAGE_SIZE?
872 */
873 slots = (KERNEL_IMAGE_SIZE - minimum - image_size) /
874 CONFIG_PHYSICAL_ALIGN + 1;
875
876 random_addr = kaslr_get_random_long("Virtual") % slots;
877
878 return random_addr * CONFIG_PHYSICAL_ALIGN + minimum;
879}
880
881/*
882 * Since this function examines addresses much more numerically,
883 * it takes the input and output pointers as 'unsigned long'.
884 */
885void choose_random_location(unsigned long input,
886 unsigned long input_size,
887 unsigned long *output,
888 unsigned long output_size,
889 unsigned long *virt_addr)
890{
891 unsigned long random_addr, min_addr;
892
893 if (cmdline_find_option_bool("nokaslr")) {
894 warn("KASLR disabled: 'nokaslr' on cmdline.");
895 return;
896 }
897
898#ifdef CONFIG_X86_5LEVEL
899 if (__read_cr4() & X86_CR4_LA57) {
900 __pgtable_l5_enabled = 1;
901 pgdir_shift = 48;
902 ptrs_per_p4d = 512;
903 }
904#endif
905
906 boot_params->hdr.loadflags |= KASLR_FLAG;
907
908 /* Prepare to add new identity pagetables on demand. */
909 initialize_identity_maps();
910
911 /* Record the various known unsafe memory ranges. */
912 mem_avoid_init(input, input_size, *output);
913
914 /*
915 * Low end of the randomization range should be the
916 * smaller of 512M or the initial kernel image
917 * location:
918 */
919 min_addr = min(*output, 512UL << 20);
920
921 /* Walk available memory entries to find a random address. */
922 random_addr = find_random_phys_addr(min_addr, output_size);
923 if (!random_addr) {
924 warn("Physical KASLR disabled: no suitable memory region!");
925 } else {
926 /* Update the new physical address location. */
927 if (*output != random_addr) {
928 add_identity_map(random_addr, output_size);
929 *output = random_addr;
930 }
931
932 /*
933 * This loads the identity mapping page table.
934 * This should only be done if a new physical address
935 * is found for the kernel, otherwise we should keep
936 * the old page table to make it be like the "nokaslr"
937 * case.
938 */
939 finalize_identity_maps();
940 }
941
942
943 /* Pick random virtual address starting from LOAD_PHYSICAL_ADDR. */
944 if (IS_ENABLED(CONFIG_X86_64))
945 random_addr = find_random_virt_addr(LOAD_PHYSICAL_ADDR, output_size);
946 *virt_addr = random_addr;
947}